{ "model_path": "/models/minimax_h3_fl2va_pruned_nvfp4.safetensors", "block_index": 24, "attention": "sage2", "warmup": 3, "iterations": 10, "profiler_iterations": 2, "quantizers": [ "comfy", "vortex_recalculate", "vortex_native" ], "metadata": { "width": 1344, "height": 768, "frames": 124, "steps": 12, "sampler_step": 1, "seed": 440420, "text_tokens": 100, "hidden_shape": [ 37810, 5376 ], "segments": [ [ 0, 100, 1 ], [ 100, 514, 2 ], [ 514, 37810, 0 ] ] }, "results": [ { "name": "attn_qkv_proj", "quantizer": "comfy", "module": { "class": "Nvfp4Linear", "in_features": 5376, "out_features": 21504, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 21504, 2688 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 21504, 336 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 5376 ], "input_is_contiguous": true, "input_stride": [ 5376, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 6.0496153309941295e-06, "p50_s": 4.536035703495145e-06, "p90_s": 9.27842338569462e-06, "p95_s": 1.3015206786803892e-05, "p99_s": 1.6004633507691324e-05, "min_s": 3.936002030968666e-06, "max_s": 1.675199018791318e-05 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 1.2190302368253469e-05, "p50_s": 9.69597022049129e-06, "p90_s": 1.4918379019945852e-05, "p95_s": 2.5099195772781942e-05, "p99_s": 3.324384917505086e-05, "min_s": 7.95203959569335e-06, "max_s": 3.5280012525618076e-05 }, "bias_cast": { "count": 10, "mean_s": 3.7487887311726807e-06, "p50_s": 3.6879791878163815e-06, "p90_s": 4.092749441042542e-06, "p95_s": 4.22238081227988e-06, "p99_s": 4.326085909269751e-06, "min_s": 3.503984771668911e-06, "max_s": 4.352012183517218e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.007912381808273495, "p50_s": 0.007679425994865596, "p90_s": 0.008507974818348885, "p95_s": 0.00887430993025191, "p99_s": 0.009167378019774334, "min_s": 0.007318171032238752, "max_s": 0.009240645042154938 }, "linear": { "count": 10, "mean_s": 0.07953937528654934, "p50_s": 0.04432033549528569, "p90_s": 0.1047707318968604, "p95_s": 0.2624061949405583, "p99_s": 0.3885145653755172, "min_s": 0.023811994993593544, "max_s": 0.42004165798425674 }, "slice_reshape": { "count": 10, "mean_s": 7.45270517654717e-06, "p50_s": 6.400019628927112e-06, "p90_s": 9.331200271844862e-06, "p95_s": 1.2369587784633033e-05, "p99_s": 1.4800297794863581e-05, "min_s": 5.6480057537555695e-06, "max_s": 1.5407975297421217e-05 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 5376 ], "input_bytes": 406533120, "packed_qdata_bytes": 101670912, "block_scale_bytes": 12708864, "output_bytes": 1626132480, "separate_materialization_bytes": 2147045376, "epilogue_candidate": "QKV views/layout and Q/K normalization/RoPE" }, "profiler_summary": { "device_kernel_launches": 0, "unique_device_kernels": 0, "positive_self_device_allocated_bytes": 4297373696, "negative_self_device_allocated_bytes": -4297373696, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/attn_qkv_proj-comfy.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::abs", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3996.745999999999, "self_cpu_time_total_us": 814.9270000000015, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813066240, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376]]" }, { "key": "aten::abs", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3158.5549999999985, "self_cpu_time_total_us": 3140.0429999999997, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813066240, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [0]]" }, { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2818.124, "self_cpu_time_total_us": 1045.0059999999976, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3253469184, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [21504, 5376], []]" }, { "key": "comfy_kitchen::quantize_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1944.1740000000027, "self_cpu_time_total_us": 756.0640000000021, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 230834176, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [], [], [], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1757.6620000000003, "self_cpu_time_total_us": 1659.4219999999987, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3253469184, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 2688], [21504, 2688], [], [], [37888, 336], [21504, 336], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1179.0390000000007, "self_cpu_time_total_us": 1172.287000000002, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1153.9020000000019, "self_cpu_time_total_us": 24.9120000000039, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 27492352, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1122.221999999998, "self_cpu_time_total_us": 1046.7179999999971, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336]]" }, { "key": "aten::amax", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 685.6779999999999, "self_cpu_time_total_us": 680.2539999999972, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[37810, 5376], [], []]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 334.9119999999966, "self_cpu_time_total_us": 17.247999999995955, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::_to_copy", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 317.66400000000067, "self_cpu_time_total_us": 208.70400000000154, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::div", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 85.58400000000256, "self_cpu_time_total_us": 85.58400000000256, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::copy_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 84.95999999999913, "self_cpu_time_total_us": 84.95999999999913, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 75.50400000000081, "self_cpu_time_total_us": 75.50400000000081, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 67.07199999999648, "self_cpu_time_total_us": 67.07199999999648, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "aten::empty", "count": 12, "device_type": "DeviceType.CPU", "cpu_time_total_us": 60.55999999999767, "self_cpu_time_total_us": 60.55999999999767, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3484303360, "self_device_memory_usage_bytes": 3484303360, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::empty_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 24.0, "self_cpu_time_total_us": 24.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::resize_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 18.511999999998807, "self_cpu_time_total_us": 18.511999999998807, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813066240, "self_device_memory_usage_bytes": 813066240, "flops": 0, "input_shapes": "[[0], [], []]" }, { "key": "aten::reshape", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 17.247999999999593, "self_cpu_time_total_us": 12.175999999999476, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 15.45600000000195, "self_cpu_time_total_us": 12.864000000005035, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 21504], [], [], [], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 15.072000000003754, "self_cpu_time_total_us": 15.072000000003754, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.7519999999985885, "self_cpu_time_total_us": 6.7519999999985885, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.424000000002707, "self_cpu_time_total_us": 5.424000000002707, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.072000000000116, "self_cpu_time_total_us": 5.072000000000116, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.600000000002183, "self_cpu_time_total_us": 3.600000000002183, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504]]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.1679999999978463, "self_cpu_time_total_us": 1.8399999999965075, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.591999999996915, "self_cpu_time_total_us": 2.591999999996915, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 21504], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.5280000000020664, "self_cpu_time_total_us": 2.5280000000020664, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[21504, 336], []]" }, { "key": "aten::to", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.2079999999969004, "self_cpu_time_total_us": 2.2079999999969004, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.3280000000013388, "self_cpu_time_total_us": 1.3280000000013388, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504], []]" }, { "key": "[memory]", "count": 18, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -4297372672, "self_device_memory_usage_bytes": -4297372672, "flops": 0, "input_shapes": "[]" } ] }, { "name": "attn_qkv_proj", "quantizer": "vortex_recalculate", "module": { "class": "Nvfp4Linear", "in_features": 5376, "out_features": 21504, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 21504, 2688 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 21504, 336 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 5376 ], "input_is_contiguous": true, "input_stride": [ 5376, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 6.164796650409698e-06, "p50_s": 5.215988494455814e-06, "p90_s": 8.035183418542146e-06, "p95_s": 8.409592555835842e-06, "p99_s": 8.7091198656708e-06, "min_s": 4.57599526271224e-06, "max_s": 8.78400169312954e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 0.00016132930177263914, "p50_s": 1.0535994078963995e-05, "p90_s": 0.00017684289487078733, "p95_s": 0.0008282039430923744, "p99_s": 0.0013492927816696467, "min_s": 8.000002708286047e-06, "max_s": 0.0014795649913139641 }, "bias_cast": { "count": 10, "mean_s": 5.227205110713839e-06, "p50_s": 4.3920299503952265e-06, "p90_s": 5.606387276202437e-06, "p95_s": 9.163195500150314e-06, "p99_s": 1.200864207930863e-05, "min_s": 4.096014890819788e-06, "max_s": 1.2720003724098206e-05 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.0016794118040706963, "p50_s": 0.0016737655096221715, "p90_s": 0.001700576167786494, "p95_s": 0.0017489025805843992, "p99_s": 0.0017875637108227239, "min_s": 0.0016442370251752436, "max_s": 0.001797228993382305 }, "scale_compat_cast": { "count": 10, "mean_s": 2.4196697631850837e-05, "p50_s": 1.6175967175513506e-05, "p90_s": 3.5485514672473066e-05, "p95_s": 4.8294765292666825e-05, "p99_s": 5.854216578882188e-05, "min_s": 1.1936004739254713e-05, "max_s": 6.110401591286063e-05 }, "scale_to_device": { "count": 10, "mean_s": 1.4910398749634624e-05, "p50_s": 1.364797935821116e-05, "p90_s": 1.9180780509486794e-05, "p95_s": 2.0174376550130544e-05, "p99_s": 2.096925338264555e-05, "min_s": 1.0959978681057692e-05, "max_s": 2.1167972590774298e-05 }, "ck_quantize_nvfp4": { "count": 10, "mean_s": 0.0032863131840713323, "p50_s": 0.002224964991910383, "p90_s": 0.0034175936831161344, "p95_s": 0.00809333081706426, "p99_s": 0.011833920524222778, "min_s": 0.0021520599839277565, "max_s": 0.012769067951012403 }, "params_wrap": { "count": 10, "mean_s": 1.0996783385053276e-05, "p50_s": 8.327973773702979e-06, "p90_s": 1.9836780847981568e-05, "p95_s": 2.3206384503282598e-05, "p99_s": 2.5902067427523434e-05, "min_s": 6.83200778439641e-06, "max_s": 2.657598815858364e-05 }, "activation_quantize": { "count": 10, "mean_s": 0.0050613199127838016, "p50_s": 0.003989738033851609, "p90_s": 0.0053149498358834495, "p95_s": 0.009906519434298378, "p99_s": 0.013579775113030339, "min_s": 0.003893512999638915, "max_s": 0.014498089032713324 }, "linear": { "count": 10, "mean_s": 0.03375178879359737, "p50_s": 0.034510504483478144, "p90_s": 0.04217453677556478, "p95_s": 0.04223313039110508, "p99_s": 0.04228000528353732, "min_s": 0.02445064898347482, "max_s": 0.04229172400664538 }, "slice_reshape": { "count": 10, "mean_s": 8.408003486692905e-06, "p50_s": 7.824011845514178e-06, "p90_s": 1.0484817903488874e-05, "p95_s": 1.08664040453732e-05, "p99_s": 1.1171672958880663e-05, "min_s": 6.607966497540474e-06, "max_s": 1.1247990187257528e-05 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 5376 ], "input_bytes": 406533120, "packed_qdata_bytes": 101670912, "block_scale_bytes": 12708864, "output_bytes": 1626132480, "separate_materialization_bytes": 2147045376, "epilogue_candidate": "QKV views/layout and Q/K normalization/RoPE" }, "profiler_summary": { "device_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 3485617152, "negative_self_device_allocated_bytes": -3485617152, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/attn_qkv_proj-vortex_recalculate.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 525.5680000000002, "self_cpu_time_total_us": 122.27200000000039, "device_time_total_us": 50787.75, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3254779904, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [21504, 5376], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 389.1359999999995, "self_cpu_time_total_us": 292.04799999999886, "device_time_total_us": 50787.75, "self_device_time_total_us": 50783.78200000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3254779904, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 2688], [21504, 2688], [], [], [37888, 336], [21504, 336], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 50783.78200000001, "self_device_time_total_us": 50783.78200000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "comfy_kitchen::quantize_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 518.3830000000003, "self_cpu_time_total_us": 398.92799999999943, "device_time_total_us": 4210.547999999997, "self_device_time_total_us": 4165.332, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 230834176, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [], [], [], []]" }, { "key": "void comfy::(anonymous namespace)::quantize_nvfp4_kernel<__nv_bfloat16, __nv_fp4x2_e2m1, __nv_fp8_e4m3, true, true>(__nv_bfloat16 const*, float const*, __nv_fp4x2_e2m1*, __nv_fp8_e4m3*, unsigned long, unsigned long, unsigned long, unsigned long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4165.332, "self_device_time_total_us": 4165.332, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::partial_absmax_bf16_vec_kernel(uint4 const*, unsigned short const*, float*, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3297.079, "self_device_time_total_us": 3297.079, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 62.128000000000156, "self_cpu_time_total_us": 16.40000000000009, "device_time_total_us": 45.21599999999671, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 27492352, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 39.952000000000226, "self_cpu_time_total_us": 10.559999999999945, "device_time_total_us": 45.21599999999671, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 29.39200000000028, "self_cpu_time_total_us": 13.232000000000426, "device_time_total_us": 45.21599999999671, "self_device_time_total_us": 45.21599999999671, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 45.21599999999671, "self_device_time_total_us": 45.21599999999671, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 188.20600000000013, "self_cpu_time_total_us": 27.840000000000146, "device_time_total_us": 5.503000000006523, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 110.94199999999955, "self_cpu_time_total_us": 69.98299999999927, "device_time_total_us": 5.503000000006523, "self_device_time_total_us": 5.503000000006523, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 39.264000000000124, "self_cpu_time_total_us": 22.463999999999942, "device_time_total_us": 3.9679999999962092, "self_device_time_total_us": 3.9679999999962092, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.9679999999962092, "self_device_time_total_us": 3.9679999999962092, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::final_scale_warp_kernel(float const*, float*, long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.871999999997115, "self_device_time_total_us": 3.871999999997115, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 36.751000000000204, "self_cpu_time_total_us": 3.088000000000193, "device_time_total_us": 2.9750000000008185, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.9750000000008185, "self_device_time_total_us": 2.9750000000008185, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 168.75099999999918, "self_cpu_time_total_us": 14.20799999999906, "device_time_total_us": 2.5280000000057044, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.5280000000057044, "self_device_time_total_us": 2.5280000000057044, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 56534.981, "self_cpu_time_total_us": 56534.981, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1254.8939999999998, "self_cpu_time_total_us": 1254.8939999999998, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 209.16600000000062, "self_cpu_time_total_us": 209.16600000000062, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 49.42400000000043, "self_cpu_time_total_us": 49.42400000000043, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 2048, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::empty", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 34.943999999999505, "self_cpu_time_total_us": 34.943999999999505, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3485614080, "self_device_memory_usage_bytes": 3485614080, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::reshape", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 18.12800000000061, "self_cpu_time_total_us": 8.896000000000186, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 17.696000000000367, "self_cpu_time_total_us": 17.696000000000367, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 16.65599999999995, "self_cpu_time_total_us": 16.65599999999995, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 14.16000000000031, "self_cpu_time_total_us": 10.176000000000386, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 21504], [], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 13.791999999999916, "self_cpu_time_total_us": 8.415999999999826, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 9.232000000000426, "self_cpu_time_total_us": 9.232000000000426, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 8.000000000000455, "self_cpu_time_total_us": 8.000000000000455, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.37600000000009, "self_cpu_time_total_us": 5.37600000000009, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.9839999999999236, "self_cpu_time_total_us": 3.9839999999999236, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 21504], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.5359999999996035, "self_cpu_time_total_us": 2.207999999999629, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.4720000000002074, "self_cpu_time_total_us": 1.4720000000002074, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.3600000000001273, "self_cpu_time_total_us": 1.3600000000001273, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[21504, 336], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.3279999999999745, "self_cpu_time_total_us": 1.3279999999999745, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504], []]" }, { "key": "[memory]", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -3485616128, "self_device_memory_usage_bytes": -3485616128, "flops": 0, "input_shapes": "[]" } ] }, { "name": "attn_qkv_proj", "quantizer": "vortex_native", "module": { "class": "Nvfp4Linear", "in_features": 5376, "out_features": 21504, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 21504, 2688 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 21504, 336 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 5376 ], "input_is_contiguous": true, "input_stride": [ 5376, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 5.945604061707854e-06, "p50_s": 5.079986294731498e-06, "p90_s": 7.5663847383111715e-06, "p95_s": 7.631213520653546e-06, "p99_s": 7.683076546527445e-06, "min_s": 4.688045009970665e-06, "max_s": 7.69604230299592e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 1.253918744623661e-05, "p50_s": 9.055976988747716e-06, "p90_s": 1.8667196854948994e-05, "p95_s": 2.426159335300325e-05, "p99_s": 2.873711055144668e-05, "min_s": 8.256000000983477e-06, "max_s": 2.985598985105753e-05 }, "bias_cast": { "count": 10, "mean_s": 4.419207107275724e-06, "p50_s": 4.344008630141616e-06, "p90_s": 4.886416718363762e-06, "p95_s": 4.915229510515928e-06, "p99_s": 4.938279744237661e-06, "min_s": 3.936002030968666e-06, "max_s": 4.944042302668095e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.001660768303554505, "p50_s": 0.0016563970129936934, "p90_s": 0.0016738977981731295, "p95_s": 0.0016784913954325021, "p99_s": 0.00168216627324, "min_s": 0.0016487810062244534, "max_s": 0.0016830849926918745 }, "scale_compat_cast": { "count": 10, "mean_s": 0.0012001517054159195, "p50_s": 1.4680001186206937e-05, "p90_s": 0.0021871404955163564, "p95_s": 0.006389521749224504, "p99_s": 0.00975142675219104, "min_s": 1.1808006092905998e-05, "max_s": 0.010591903002932668 }, "scale_to_device": { "count": 10, "mean_s": 2.060000551864505e-05, "p50_s": 1.736002741381526e-05, "p90_s": 3.288962761871517e-05, "p95_s": 3.667680721264331e-05, "p99_s": 3.970655088778585e-05, "min_s": 1.0911957360804081e-05, "max_s": 4.0463986806571484e-05 }, "vortex_quantize_nvfp4": { "count": 10, "mean_s": 0.0022807597066275775, "p50_s": 0.0022093004954513162, "p90_s": 0.002431947988225147, "p95_s": 0.002448867986095138, "p99_s": 0.0024624039843911304, "min_s": 0.002181821037083864, "max_s": 0.0024657879839651287 }, "params_wrap": { "count": 10, "mean_s": 1.1233595432713628e-05, "p50_s": 7.1919639594852924e-06, "p90_s": 2.6147201424464583e-05, "p95_s": 2.702559868339449e-05, "p99_s": 2.7728316490538417e-05, "min_s": 6.783986464142799e-06, "max_s": 2.79039959423244e-05 }, "activation_quantize": { "count": 10, "mean_s": 0.005210591707145796, "p50_s": 0.004148697014898062, "p90_s": 0.006150649720802899, "p95_s": 0.010348617378622284, "p99_s": 0.013706991504877808, "min_s": 0.0038902339874766767, "max_s": 0.014546585036441684 }, "linear": { "count": 10, "mean_s": 0.026462141395313665, "p50_s": 0.025728374515892938, "p90_s": 0.027108474890701468, "p95_s": 0.03137550492538138, "p99_s": 0.034789128953125324, "min_s": 0.024055432993918657, "max_s": 0.03564253496006131 }, "slice_reshape": { "count": 10, "mean_s": 7.641606498509645e-06, "p50_s": 7.071997970342636e-06, "p90_s": 9.409559424966574e-06, "p95_s": 9.77679155766964e-06, "p99_s": 1.0070577263832093e-05, "min_s": 6.384041626006365e-06, "max_s": 1.0144023690372705e-05 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 5376 ], "input_bytes": 406533120, "packed_qdata_bytes": 101670912, "block_scale_bytes": 12708864, "output_bytes": 1626132480, "separate_materialization_bytes": 2147045376, "epilogue_candidate": "QKV views/layout and Q/K normalization/RoPE" }, "profiler_summary": { "device_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 3484306432, "negative_self_device_allocated_bytes": -3484306432, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/attn_qkv_proj-vortex_native.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 641.3119999999994, "self_cpu_time_total_us": 113.00799999999936, "device_time_total_us": 51407.981999999996, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3253469184, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [21504, 5376], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 517.6480000000001, "self_cpu_time_total_us": 422.4639999999981, "device_time_total_us": 51407.981999999996, "self_device_time_total_us": 51404.11, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3253469184, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 2688], [21504, 2688], [], [], [37888, 336], [21504, 336], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 51404.11, "self_device_time_total_us": 51404.11, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::quantize_nvfp4_bf16_kernel(unsigned short const*, float const*, unsigned char*, unsigned char*, long, long, long, long, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4352.473000000004, "self_device_time_total_us": 4352.473000000004, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::partial_absmax_bf16_vec_kernel(uint4 const*, unsigned short const*, float*, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3293.211000000001, "self_device_time_total_us": 3293.211000000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 58.43199999999979, "self_cpu_time_total_us": 11.695999999999458, "device_time_total_us": 45.59900000000107, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 27492352, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 41.76000000000022, "self_cpu_time_total_us": 13.728000000000065, "device_time_total_us": 45.59900000000107, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 28.032000000000153, "self_cpu_time_total_us": 13.071999999999662, "device_time_total_us": 45.59900000000107, "self_device_time_total_us": 45.59900000000107, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 45.59900000000107, "self_device_time_total_us": 45.59900000000107, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 184.27199999999948, "self_cpu_time_total_us": 35.91999999999939, "device_time_total_us": 5.5359999999996035, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 80.25599999999986, "self_cpu_time_total_us": 45.23199999999906, "device_time_total_us": 5.5359999999996035, "self_device_time_total_us": 5.5359999999996035, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "(anonymous namespace)::final_scale_warp_kernel(float const*, float*, long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4.000000000004093, "self_device_time_total_us": 4.000000000004093, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 36.48000000000002, "self_cpu_time_total_us": 20.864000000000487, "device_time_total_us": 3.87199999999757, "self_device_time_total_us": 3.87199999999757, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.87199999999757, "self_device_time_total_us": 3.87199999999757, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 38.81600000000026, "self_cpu_time_total_us": 2.9280000000003383, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 2.9760000000005675, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 162.99199999999917, "self_cpu_time_total_us": 14.607999999999606, "device_time_total_us": 2.559999999999036, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.559999999999036, "self_device_time_total_us": 2.559999999999036, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 57783.187, "self_cpu_time_total_us": 57783.187, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1088.8139999999999, "self_cpu_time_total_us": 1088.8139999999999, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 162.4160000000004, "self_cpu_time_total_us": 162.4160000000004, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 68.09600000000023, "self_cpu_time_total_us": 68.09600000000023, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 2048, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::empty", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 35.36000000000058, "self_cpu_time_total_us": 35.36000000000058, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 3484303360, "self_device_memory_usage_bytes": 3484303360, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 17.50400000000036, "self_cpu_time_total_us": 17.50400000000036, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 15.024000000000342, "self_cpu_time_total_us": 15.024000000000342, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 13.376000000000317, "self_cpu_time_total_us": 8.1760000000003, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 10.655999999999949, "self_cpu_time_total_us": 7.983999999999924, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 21504], [], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 9.135999999999967, "self_cpu_time_total_us": 5.8400000000001455, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.176000000000386, "self_cpu_time_total_us": 6.176000000000386, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.200000000000017, "self_cpu_time_total_us": 5.200000000000017, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.7599999999997635, "self_cpu_time_total_us": 2.6559999999994943, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.2959999999998217, "self_cpu_time_total_us": 3.2959999999998217, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.6720000000000255, "self_cpu_time_total_us": 2.6720000000000255, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 21504], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.6960000000003674, "self_cpu_time_total_us": 1.6960000000003674, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[21504, 336], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.6640000000002146, "self_cpu_time_total_us": 1.6640000000002146, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.1040000000002692, "self_cpu_time_total_us": 1.1040000000002692, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 21504], []]" }, { "key": "[memory]", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -3484305408, "self_device_memory_usage_bytes": -3484305408, "flops": 0, "input_shapes": "[]" } ] }, { "name": "attn_out_proj", "quantizer": "comfy", "module": { "class": "Nvfp4Linear", "in_features": 7168, "out_features": 5376, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 5376, 3584 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 5376, 448 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 7168 ], "input_is_contiguous": true, "input_stride": [ 7168, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 4.9375870730727914e-06, "p50_s": 4.639994585886598e-06, "p90_s": 5.185569170862435e-06, "p95_s": 6.4167659729719135e-06, "p99_s": 7.4017234146595e-06, "min_s": 4.496017936617136e-06, "max_s": 7.647962775081396e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 7.991993334144354e-06, "p50_s": 7.855996955186129e-06, "p90_s": 8.633575635030866e-06, "p95_s": 8.820780203677713e-06, "p99_s": 8.970543858595192e-06, "min_s": 7.520022336393595e-06, "max_s": 9.007984772324562e-06 }, "bias_cast": { "count": 10, "mean_s": 4.151801113039255e-06, "p50_s": 4.120025550946593e-06, "p90_s": 4.21190052293241e-06, "p95_s": 4.3779407860711215e-06, "p99_s": 4.510772996582091e-06, "min_s": 4.047004040330648e-06, "max_s": 4.543981049209833e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.009642405400518328, "p50_s": 0.009569192508934066, "p90_s": 0.009928448009304702, "p95_s": 0.010009808000177145, "p99_s": 0.010074895992875098, "min_s": 0.009391825005877763, "max_s": 0.010091167991049588 }, "linear": { "count": 10, "mean_s": 0.008343223499832676, "p50_s": 0.008321458473801613, "p90_s": 0.008729003596818075, "p95_s": 0.008753238795907236, "p99_s": 0.008772626955178565, "min_s": 0.007807716028764844, "max_s": 0.008777473994996399 }, "slice_reshape": { "count": 10, "mean_s": 6.433599628508091e-06, "p50_s": 6.351998308673501e-06, "p90_s": 6.64321705698967e-06, "p95_s": 6.873614620417356e-06, "p99_s": 7.057932671159506e-06, "min_s": 6.143993232399225e-06, "max_s": 7.104012183845043e-06 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 7168 ], "input_bytes": 542044160, "packed_qdata_bytes": 135561216, "block_scale_bytes": 16945152, "output_bytes": 406533120, "separate_materialization_bytes": 1101083648, "epilogue_candidate": "segmented BF16 residual gate/add" }, "profiler_summary": { "device_kernel_launches": 18, "unique_device_kernels": 9, "positive_self_device_allocated_bytes": 2202529792, "negative_self_device_allocated_bytes": -2202529792, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/attn_out_proj-comfy.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 693.0239999999999, "self_cpu_time_total_us": 138.5279999999998, "device_time_total_us": 17302.301, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], [5376, 7168], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 547.3919999999998, "self_cpu_time_total_us": 401.4880000000003, "device_time_total_us": 17302.301, "self_device_time_total_us": 17298.301, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 3584], [5376, 3584], [], [], [37888, 448], [5376, 448], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 17298.301, "self_device_time_total_us": 17298.301, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::abs", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1081.39, "self_cpu_time_total_us": 9.600000000000364, "device_time_total_us": 13704.701000000001, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1084088320, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168]]" }, { "key": "aten::abs", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1052.1739999999998, "self_cpu_time_total_us": 43.455999999999676, "device_time_total_us": 13704.701000000001, "self_device_time_total_us": 9117.385, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1084088320, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], [0]]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::AbsFunctor, std::array >(int, at::native::AbsFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 9117.385, "self_device_time_total_us": 9117.385, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "comfy_kitchen::quantize_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 592.575, "self_cpu_time_total_us": 478.4950000000001, "device_time_total_us": 5686.86, "self_device_time_total_us": 5609.068, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 305070080, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], [], [], [], []]" }, { "key": "void comfy::(anonymous namespace)::quantize_nvfp4_kernel<__nv_bfloat16, __nv_fp4x2_e2m1, __nv_fp8_e4m3, true, true>(__nv_bfloat16 const*, float const*, __nv_fp4x2_e2m1*, __nv_fp8_e4m3*, unsigned long, unsigned long, unsigned long, unsigned long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 5609.068, "self_device_time_total_us": 5609.068, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 946.062, "self_cpu_time_total_us": 946.062, "device_time_total_us": 4587.316000000001, "self_device_time_total_us": 4587.316000000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::amax", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 179.82400000000007, "self_cpu_time_total_us": 84.25600000000009, "device_time_total_us": 4413.910999999998, "self_device_time_total_us": 4413.910999999998, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[37810, 7168], [], []]" }, { "key": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp >, unsigned int, c10::BFloat16, 4, 4> >(at::native::ReduceOp >, unsigned int, c10::BFloat16, 4, 4>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4413.046999999997, "self_device_time_total_us": 4413.046999999997, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 65.2159999999999, "self_cpu_time_total_us": 13.135999999999967, "device_time_total_us": 77.79199999999946, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 33947648, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 46.83199999999988, "self_cpu_time_total_us": 9.375999999999749, "device_time_total_us": 77.79199999999946, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 37.45600000000013, "self_cpu_time_total_us": 15.10400000000027, "device_time_total_us": 77.79199999999946, "self_device_time_total_us": 77.79199999999946, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 77.79199999999946, "self_device_time_total_us": 77.79199999999946, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 82.12800000000016, "self_cpu_time_total_us": 25.184000000000196, "device_time_total_us": 4.0, "self_device_time_total_us": 4.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4.0, "self_device_time_total_us": 4.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::div", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 69.36000000000013, "self_cpu_time_total_us": 43.96799999999985, "device_time_total_us": 3.7119999999995343, "self_device_time_total_us": 3.7119999999995343, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BUnaryFunctor >, std::array >(int, at::native::BUnaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.7119999999995343, "self_device_time_total_us": 3.7119999999995343, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 105.87200000000007, "self_cpu_time_total_us": 10.192000000000235, "device_time_total_us": 2.4629999999988286, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::_to_copy", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 95.67999999999984, "self_cpu_time_total_us": 21.695999999999913, "device_time_total_us": 2.4629999999988286, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 63.67999999999961, "self_cpu_time_total_us": 34.991999999999734, "device_time_total_us": 2.4629999999988286, "self_device_time_total_us": 2.4629999999988286, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.4629999999988286, "self_device_time_total_us": 2.4629999999988286, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Memset (Device)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.864000000001397, "self_device_time_total_us": 0.864000000001397, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 34483.321, "self_cpu_time_total_us": 34483.321, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 227.69599999999969, "self_cpu_time_total_us": 227.69599999999969, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaMemsetAsync", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 61.4079999999999, "self_cpu_time_total_us": 61.4079999999999, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty", "count": 12, "device_type": "DeviceType.CPU", "cpu_time_total_us": 46.09600000000046, "self_cpu_time_total_us": 46.09600000000046, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1118437376, "self_device_memory_usage_bytes": 1118437376, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 24.032000000000153, "self_cpu_time_total_us": 24.032000000000153, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::resize_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 23.408000000000072, "self_cpu_time_total_us": 23.408000000000072, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1084088320, "self_device_memory_usage_bytes": 1084088320, "flops": 0, "input_shapes": "[[0], [], []]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 21.55199999999968, "self_cpu_time_total_us": 21.55199999999968, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 13.760000000000076, "self_cpu_time_total_us": 8.65600000000029, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], []]" }, { "key": "aten::empty_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 10.304000000000315, "self_cpu_time_total_us": 10.304000000000315, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 8.399999999999181, "self_cpu_time_total_us": 8.399999999999181, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448], []]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 7.104000000000269, "self_cpu_time_total_us": 5.871999999999844, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], [], []]" }, { "key": "aten::reshape", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.816000000000258, "self_cpu_time_total_us": 4.656000000000404, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.736000000000104, "self_cpu_time_total_us": 6.736000000000104, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.103999999999786, "self_cpu_time_total_us": 5.103999999999786, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.119999999999891, "self_cpu_time_total_us": 2.4320000000002437, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.9600000000000364, "self_cpu_time_total_us": 2.9600000000000364, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[5376, 448], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.1599999999998545, "self_cpu_time_total_us": 2.1599999999998545, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.6640000000002146, "self_cpu_time_total_us": 1.6640000000002146, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376]]" }, { "key": "aten::to", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.536000000000115, "self_cpu_time_total_us": 1.536000000000115, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.2320000000004256, "self_cpu_time_total_us": 1.2320000000004256, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.6879999999996471, "self_cpu_time_total_us": 0.6879999999996471, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "[memory]", "count": 18, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -2202528768, "self_device_memory_usage_bytes": -2202528768, "flops": 0, "input_shapes": "[]" } ] }, { "name": "attn_out_proj", "quantizer": "vortex_recalculate", "module": { "class": "Nvfp4Linear", "in_features": 7168, "out_features": 5376, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 5376, 3584 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 5376, 448 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 7168 ], "input_is_contiguous": true, "input_stride": [ 7168, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 6.518402369692922e-06, "p50_s": 4.640023689717054e-06, "p90_s": 8.460803655907508e-06, "p95_s": 1.348640944343059e-05, "p99_s": 1.7506894073449077e-05, "min_s": 4.431989509612322e-06, "max_s": 1.8512015230953693e-05 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 8.603202877566218e-06, "p50_s": 8.216011337935925e-06, "p90_s": 9.257608326151965e-06, "p95_s": 1.0740812285803255e-05, "p99_s": 1.1927375453524292e-05, "min_s": 7.679976988583803e-06, "max_s": 1.222401624545455e-05 }, "bias_cast": { "count": 10, "mean_s": 4.187191370874643e-06, "p50_s": 4.10401844419539e-06, "p90_s": 4.600000102072954e-06, "p95_s": 4.635989898815752e-06, "p99_s": 4.664781736209988e-06, "min_s": 3.96795803681016e-06, "max_s": 4.671979695558548e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.0021354953991249205, "p50_s": 0.0021257479675114155, "p90_s": 0.0021507832396309825, "p95_s": 0.0021839896420715378, "p99_s": 0.002210554764023982, "min_s": 0.002113804977852851, "max_s": 0.002217196044512093 }, "scale_compat_cast": { "count": 10, "mean_s": 1.7780804773792626e-05, "p50_s": 1.193603384308517e-05, "p90_s": 1.879042829386888e-05, "p95_s": 4.351523530203843e-05, "p99_s": 6.329508090857418e-05, "min_s": 1.1280004400759935e-05, "max_s": 6.824004231020808e-05 }, "scale_to_device": { "count": 10, "mean_s": 1.2153509305790067e-05, "p50_s": 1.0727497283369303e-05, "p90_s": 1.528642023913562e-05, "p95_s": 1.841121993493288e-05, "p99_s": 2.09110596915707e-05, "min_s": 1.0400020983070135e-05, "max_s": 2.1536019630730152e-05 }, "ck_quantize_nvfp4": { "count": 10, "mean_s": 0.0028809011972043665, "p50_s": 0.0028607480053324252, "p90_s": 0.002952268690569326, "p95_s": 0.002984316347283311, "p99_s": 0.0030099544726544993, "min_s": 0.00283020397182554, "max_s": 0.003016364003997296 }, "params_wrap": { "count": 10, "mean_s": 8.480006363242864e-06, "p50_s": 6.720016244798899e-06, "p90_s": 9.32640978135168e-06, "p95_s": 1.637519744690506e-05, "p99_s": 2.2014227579347792e-05, "min_s": 6.4639607444405556e-06, "max_s": 2.3423985112458467e-05 }, "activation_quantize": { "count": 10, "mean_s": 0.005091450706822797, "p50_s": 0.005055840010754764, "p90_s": 0.005223735910840332, "p95_s": 0.005299623461905867, "p99_s": 0.005360333502758294, "min_s": 0.005004151025786996, "max_s": 0.005375511012971401 }, "linear": { "count": 10, "mean_s": 0.008391089201904834, "p50_s": 0.0083667790167965, "p90_s": 0.008696587570011616, "p95_s": 0.008707358787069098, "p99_s": 0.008715975760715082, "min_s": 0.008024482987821102, "max_s": 0.008718130004126579 }, "slice_reshape": { "count": 10, "mean_s": 6.70079025439918e-06, "p50_s": 6.407994078472257e-06, "p90_s": 7.02879624441266e-06, "p95_s": 8.058408275246618e-06, "p99_s": 8.882097899913788e-06, "min_s": 6.239977665245533e-06, "max_s": 9.08802030608058e-06 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 7168 ], "input_bytes": 542044160, "packed_qdata_bytes": 135561216, "block_scale_bytes": 16945152, "output_bytes": 406533120, "separate_materialization_bytes": 1101083648, "epilogue_candidate": "segmented BF16 residual gate/add" }, "profiler_summary": { "device_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 1120353280, "negative_self_device_allocated_bytes": -1120353280, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/attn_out_proj-vortex_recalculate.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 314.89499999999987, "self_cpu_time_total_us": 56.24000000000001, "device_time_total_us": 16426.848, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 815280128, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], [5376, 7168], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 250.41499999999985, "self_cpu_time_total_us": 190.55899999999997, "device_time_total_us": 16426.848, "self_device_time_total_us": 16423.04, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 815280128, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 3584], [5376, 3584], [], [], [37888, 448], [5376, 448], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 16423.04, "self_device_time_total_us": 16423.04, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "comfy_kitchen::quantize_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 257.75900000000024, "self_cpu_time_total_us": 188.3510000000001, "device_time_total_us": 5623.03, "self_device_time_total_us": 5545.046, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 305070080, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], [], [], [], []]" }, { "key": "void comfy::(anonymous namespace)::quantize_nvfp4_kernel<__nv_bfloat16, __nv_fp4x2_e2m1, __nv_fp8_e4m3, true, true>(__nv_bfloat16 const*, float const*, __nv_fp4x2_e2m1*, __nv_fp8_e4m3*, unsigned long, unsigned long, unsigned long, unsigned long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 5545.046, "self_device_time_total_us": 5545.046, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::partial_absmax_bf16_vec_kernel(uint4 const*, unsigned short const*, float*, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4233.208000000001, "self_device_time_total_us": 4233.208000000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 42.25599999999997, "self_cpu_time_total_us": 6.33600000000024, "device_time_total_us": 77.98399999999947, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 33947648, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 30.86399999999992, "self_cpu_time_total_us": 11.935999999999922, "device_time_total_us": 77.98399999999947, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 18.927999999999997, "self_cpu_time_total_us": 7.295999999999935, "device_time_total_us": 77.98399999999947, "self_device_time_total_us": 77.98399999999947, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 77.98399999999947, "self_device_time_total_us": 77.98399999999947, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 157.34400000000016, "self_cpu_time_total_us": 22.51199999999983, "device_time_total_us": 5.503999999998541, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 97.58400000000023, "self_cpu_time_total_us": 75.34400000000045, "device_time_total_us": 5.503999999998541, "self_device_time_total_us": 5.503999999998541, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "(anonymous namespace)::final_scale_warp_kernel(float const*, float*, long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.903999999999087, "self_device_time_total_us": 3.903999999999087, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 25.104000000000042, "self_cpu_time_total_us": 13.984000000000037, "device_time_total_us": 3.8080000000018117, "self_device_time_total_us": 3.8080000000018117, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.8080000000018117, "self_device_time_total_us": 3.8080000000018117, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 26.65600000000012, "self_cpu_time_total_us": 1.9040000000000532, "device_time_total_us": 3.0399999999999636, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.0399999999999636, "self_device_time_total_us": 3.0399999999999636, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 140.36800000000025, "self_cpu_time_total_us": 7.776000000000153, "device_time_total_us": 2.4639999999985776, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.4639999999985776, "self_device_time_total_us": 2.4639999999985776, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 25379.704, "self_cpu_time_total_us": 25379.704, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 106.83200000000016, "self_cpu_time_total_us": 106.83200000000016, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 62.54400000000001, "self_cpu_time_total_us": 62.54400000000001, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 37.248000000000104, "self_cpu_time_total_us": 37.248000000000104, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 2048, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::empty", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 25.43999999999994, "self_cpu_time_total_us": 25.43999999999994, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1120350208, "self_device_memory_usage_bytes": 1120350208, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 11.727999999999952, "self_cpu_time_total_us": 7.231999999999971, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 11.103999999999928, "self_cpu_time_total_us": 11.103999999999928, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 8.240000000000009, "self_cpu_time_total_us": 6.336000000000013, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], [], []]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 7.279999999999973, "self_cpu_time_total_us": 7.279999999999973, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.223999999999819, "self_cpu_time_total_us": 4.30399999999986, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 4.495999999999981, "self_cpu_time_total_us": 4.495999999999981, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 4.400000000000091, "self_cpu_time_total_us": 4.400000000000091, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.3840000000000146, "self_cpu_time_total_us": 1.6479999999999109, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.919999999999959, "self_cpu_time_total_us": 1.919999999999959, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.9039999999999964, "self_cpu_time_total_us": 1.9039999999999964, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.1359999999999673, "self_cpu_time_total_us": 1.1359999999999673, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[5376, 448], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.928000000000111, "self_cpu_time_total_us": 0.928000000000111, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.7360000000001037, "self_cpu_time_total_us": 0.7360000000001037, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "[memory]", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -1120352256, "self_device_memory_usage_bytes": -1120352256, "flops": 0, "input_shapes": "[]" } ] }, { "name": "attn_out_proj", "quantizer": "vortex_native", "module": { "class": "Nvfp4Linear", "in_features": 7168, "out_features": 5376, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 5376, 3584 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 5376, 448 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 7168 ], "input_is_contiguous": true, "input_stride": [ 7168, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 5.027191946282983e-06, "p50_s": 4.735979018732905e-06, "p90_s": 5.1952141802757965e-06, "p95_s": 6.613615551032122e-06, "p99_s": 7.748336647637189e-06, "min_s": 4.4800108298659325e-06, "max_s": 8.032016921788454e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 8.164806058630347e-06, "p50_s": 7.89598561823368e-06, "p90_s": 8.761591743677855e-06, "p95_s": 9.38080484047532e-06, "p99_s": 9.876175317913293e-06, "min_s": 7.744005415588617e-06, "max_s": 1.0000017937272787e-05 }, "bias_cast": { "count": 10, "mean_s": 4.150398308411241e-06, "p50_s": 4.1199964471161366e-06, "p90_s": 4.235201049596071e-06, "p95_s": 4.3575768359005455e-06, "p99_s": 4.4554774649441245e-06, "min_s": 4.01597935706377e-06, "max_s": 4.479952622205019e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.002151049306849018, "p50_s": 0.0021311805176083, "p90_s": 0.0021834448620211334, "p95_s": 0.0022133604245027525, "p99_s": 0.002237292874488048, "min_s": 0.0021157090086489916, "max_s": 0.002243275986984372 }, "scale_compat_cast": { "count": 10, "mean_s": 1.2465595500543713e-05, "p50_s": 1.2080010492354631e-05, "p90_s": 1.4244805788621306e-05, "p95_s": 1.4554386143572627e-05, "p99_s": 1.4802050427533687e-05, "min_s": 1.1231983080506325e-05, "max_s": 1.486396649852395e-05 }, "scale_to_device": { "count": 10, "mean_s": 1.192319905385375e-05, "p50_s": 1.0927993571385741e-05, "p90_s": 1.4393642777577042e-05, "p95_s": 1.616481749806553e-05, "p99_s": 1.758175727445632e-05, "min_s": 1.0415969882160425e-05, "max_s": 1.793599221855402e-05 }, "vortex_quantize_nvfp4": { "count": 10, "mean_s": 0.0029934655001852663, "p50_s": 0.002858611987903714, "p90_s": 0.0032465396041516216, "p95_s": 0.003295938795781694, "p99_s": 0.0033354581490857528, "min_s": 0.0028280760161578655, "max_s": 0.0033453379874117672 }, "params_wrap": { "count": 10, "mean_s": 6.678409408777952e-06, "p50_s": 6.6159991547465324e-06, "p90_s": 6.9615547545254225e-06, "p95_s": 6.968784146010876e-06, "p99_s": 6.974567659199238e-06, "min_s": 6.384041626006365e-06, "max_s": 6.976013537496328e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.005205234803725034, "p50_s": 0.005049952014815062, "p90_s": 0.0054687846451997755, "p95_s": 0.005527795810485258, "p99_s": 0.0055750047427136455, "min_s": 0.005013736023101956, "max_s": 0.005586806975770742 }, "linear": { "count": 10, "mean_s": 0.008270908409031108, "p50_s": 0.008207883016439155, "p90_s": 0.00870191397261806, "p95_s": 0.008760846007498912, "p99_s": 0.008807991635403597, "min_s": 0.007719588000327349, "max_s": 0.008819778042379767 }, "slice_reshape": { "count": 10, "mean_s": 6.51999143883586e-06, "p50_s": 6.471993401646614e-06, "p90_s": 6.943999323993921e-06, "p95_s": 6.943999323993921e-06, "p99_s": 6.943999323993921e-06, "min_s": 6.2239705584943295e-06, "max_s": 6.943999323993921e-06 } }, "reference_diff": { "max": 1408.0, "mean": 51.06270217895508 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 7168 ], "input_bytes": 542044160, "packed_qdata_bytes": 135561216, "block_scale_bytes": 16945152, "output_bytes": 406533120, "separate_materialization_bytes": 1101083648, "epilogue_candidate": "segmented BF16 residual gate/add" }, "profiler_summary": { "device_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 1118440448, "negative_self_device_allocated_bytes": -1118440448, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/attn_out_proj-vortex_native.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 696.5589999999997, "self_cpu_time_total_us": 127.80799999999988, "device_time_total_us": 16839.242999999995, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], [5376, 7168], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 554.9269999999998, "self_cpu_time_total_us": 438.75199999999984, "device_time_total_us": 16839.242999999995, "self_device_time_total_us": 16835.371, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 3584], [5376, 3584], [], [], [37888, 448], [5376, 448], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 16835.371, "self_device_time_total_us": 16835.371, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::quantize_nvfp4_bf16_kernel(unsigned short const*, float const*, unsigned char*, unsigned char*, long, long, long, long, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 5570.564, "self_device_time_total_us": 5570.564, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::partial_absmax_bf16_vec_kernel(uint4 const*, unsigned short const*, float*, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4346.603000000001, "self_device_time_total_us": 4346.603000000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 49.80799999999988, "self_cpu_time_total_us": 9.055999999999813, "device_time_total_us": 77.72800000000143, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 33947648, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 36.320000000000164, "self_cpu_time_total_us": 10.368000000000166, "device_time_total_us": 77.72800000000143, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 25.951999999999998, "self_cpu_time_total_us": 13.023999999999887, "device_time_total_us": 77.72800000000143, "self_device_time_total_us": 77.72800000000143, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 77.72800000000143, "self_device_time_total_us": 77.72800000000143, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 172.28800000000012, "self_cpu_time_total_us": 24.944000000000074, "device_time_total_us": 5.503999999999905, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 97.18400000000008, "self_cpu_time_total_us": 66.70400000000018, "device_time_total_us": 5.503999999999905, "self_device_time_total_us": 5.503999999999905, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "(anonymous namespace)::final_scale_warp_kernel(float const*, float*, long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.904000000000906, "self_device_time_total_us": 3.904000000000906, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 34.38300000000004, "self_cpu_time_total_us": 20.75200000000018, "device_time_total_us": 3.87199999999757, "self_device_time_total_us": 3.87199999999757, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.87199999999757, "self_device_time_total_us": 3.87199999999757, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 32.624000000000024, "self_cpu_time_total_us": 2.7519999999999527, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 2.9760000000005675, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 151.61599999999987, "self_cpu_time_total_us": 9.199999999999818, "device_time_total_us": 2.527999999999338, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.527999999999338, "self_device_time_total_us": 2.527999999999338, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 25651.799, "self_cpu_time_total_us": 25651.799, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 116.63799999999958, "self_cpu_time_total_us": 116.63799999999958, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 64.56, "self_cpu_time_total_us": 64.56, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 50.15999999999997, "self_cpu_time_total_us": 50.15999999999997, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 2048, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 44.35200000000009, "self_cpu_time_total_us": 44.35200000000009, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 32.27199999999971, "self_cpu_time_total_us": 32.27199999999971, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1118437376, "self_device_memory_usage_bytes": 1118437376, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 16.687999999999874, "self_cpu_time_total_us": 16.687999999999874, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 14.496000000000066, "self_cpu_time_total_us": 8.448000000000178, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], []]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 13.82400000000007, "self_cpu_time_total_us": 9.072000000000116, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.384000000000242, "self_cpu_time_total_us": 2.576000000000249, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.047999999999888, "self_cpu_time_total_us": 6.047999999999888, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 7168], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 4.751999999999953, "self_cpu_time_total_us": 4.751999999999953, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.8079999999999927, "self_cpu_time_total_us": 3.8079999999999927, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.6960000000000264, "self_cpu_time_total_us": 3.6960000000000264, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 448], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.0399999999999636, "self_cpu_time_total_us": 1.9200000000000728, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.6320000000002892, "self_cpu_time_total_us": 1.6320000000002892, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.1520000000000437, "self_cpu_time_total_us": 1.1520000000000437, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[5376, 448], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.1199999999998909, "self_cpu_time_total_us": 1.1199999999998909, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "[memory]", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -1118439424, "self_device_memory_usage_bytes": -1118439424, "flops": 0, "input_shapes": "[]" } ] }, { "name": "mlp_fc1", "quantizer": "comfy", "module": { "class": "Nvfp4Linear", "in_features": 5376, "out_features": 28672, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 28672, 2688 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 28672, 336 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 5376 ], "input_is_contiguous": true, "input_stride": [ 5376, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 4.872021963819861e-06, "p50_s": 4.6000059228390455e-06, "p90_s": 5.0144211854785676e-06, "p95_s": 6.123216007836161e-06, "p99_s": 7.010251865722239e-06, "min_s": 4.496017936617136e-06, "max_s": 7.232010830193758e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 8.105597225949169e-06, "p50_s": 7.927999831736088e-06, "p90_s": 8.56962869875133e-06, "p95_s": 9.044804028235375e-06, "p99_s": 9.424944291822614e-06, "min_s": 7.744005415588617e-06, "max_s": 9.519979357719421e-06 }, "bias_cast": { "count": 10, "mean_s": 4.192005144432187e-06, "p50_s": 4.1120219975709915e-06, "p90_s": 4.332786193117499e-06, "p95_s": 4.53439715784043e-06, "p99_s": 4.695685929618775e-06, "min_s": 4.047993570566177e-06, "max_s": 4.736008122563362e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.007388551312033087, "p50_s": 0.007319628493860364, "p90_s": 0.007553186418954283, "p95_s": 0.007642171217594295, "p99_s": 0.007713359056506306, "min_s": 0.007295236049685627, "max_s": 0.0077311560162343085 }, "linear": { "count": 10, "mean_s": 0.03401212958851829, "p50_s": 0.03394539398141205, "p90_s": 0.03457325681229122, "p95_s": 0.03510591190715786, "p99_s": 0.03553203598305117, "min_s": 0.03306085895746946, "max_s": 0.0356385670020245 }, "slice_reshape": { "count": 10, "mean_s": 6.496004061773419e-06, "p50_s": 6.440008291974664e-06, "p90_s": 6.731203757226467e-06, "p95_s": 6.925611523911356e-06, "p99_s": 7.081137737259269e-06, "min_s": 6.127986125648022e-06, "max_s": 7.120019290596247e-06 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 5376 ], "input_bytes": 406533120, "packed_qdata_bytes": 101670912, "block_scale_bytes": 12708864, "output_bytes": 2168176640, "separate_materialization_bytes": 2689089536, "epilogue_candidate": "BF16 split and SwiGLU input" }, "profiler_summary": { "device_kernel_launches": 18, "unique_device_kernels": 9, "positive_self_device_allocated_bytes": 5381863424, "negative_self_device_allocated_bytes": -5381863424, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/mlp_fc1-comfy.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1490.0310000000009, "self_cpu_time_total_us": 347.7600000000002, "device_time_total_us": 67700.774, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4337958912, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [28672, 5376], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1135.6949999999997, "self_cpu_time_total_us": 1005.2789999999968, "device_time_total_us": 67700.774, "self_device_time_total_us": 67696.902, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4337958912, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 2688], [28672, 2688], [], [], [37888, 336], [28672, 336], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 67696.902, "self_device_time_total_us": 67696.902, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::abs", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2874.414999999999, "self_cpu_time_total_us": 606.847999999999, "device_time_total_us": 10481.243000000002, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813066240, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376]]" }, { "key": "aten::abs", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2247.343, "self_cpu_time_total_us": 731.3599999999997, "device_time_total_us": 10481.243000000002, "self_device_time_total_us": 6973.885000000002, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813066240, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [0]]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::AbsFunctor, std::array >(int, at::native::AbsFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 6973.885000000002, "self_device_time_total_us": 6973.885000000002, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "comfy_kitchen::quantize_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1524.686999999999, "self_cpu_time_total_us": 476.7679999999991, "device_time_total_us": 4195.678000000007, "self_device_time_total_us": 4150.206000000004, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 230834176, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [], [], [], []]" }, { "key": "void comfy::(anonymous namespace)::quantize_nvfp4_kernel<__nv_bfloat16, __nv_fp4x2_e2m1, __nv_fp8_e4m3, true, true>(__nv_bfloat16 const*, float const*, __nv_fp4x2_e2m1*, __nv_fp8_e4m3*, unsigned long, unsigned long, unsigned long, unsigned long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4150.206000000004, "self_device_time_total_us": 4150.206000000004, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 829.2470000000003, "self_cpu_time_total_us": 829.2470000000003, "device_time_total_us": 3507.358, "self_device_time_total_us": 3507.358, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::amax", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 165.39200000000073, "self_cpu_time_total_us": 74.08000000000084, "device_time_total_us": 3495.5510000000013, "self_device_time_total_us": 3495.5510000000013, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[37810, 5376], [], []]" }, { "key": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp >, unsigned int, c10::BFloat16, 4, 4> >(at::native::ReduceOp >, unsigned int, c10::BFloat16, 4, 4>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3494.6870000000035, "self_device_time_total_us": 3494.6870000000035, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 991.9030000000002, "self_cpu_time_total_us": 14.78399999999965, "device_time_total_us": 45.47200000000339, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 27492352, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 971.103000000001, "self_cpu_time_total_us": 916.6550000000007, "device_time_total_us": 45.47200000000339, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 54.44800000000032, "self_cpu_time_total_us": 19.360000000000582, "device_time_total_us": 45.47200000000339, "self_device_time_total_us": 45.47200000000339, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 45.47200000000339, "self_device_time_total_us": 45.47200000000339, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 47.823999999998705, "self_cpu_time_total_us": 20.89599999999882, "device_time_total_us": 3.872000000004846, "self_device_time_total_us": 3.872000000004846, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.872000000004846, "self_device_time_total_us": 3.872000000004846, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::div", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 50.25600000000031, "self_cpu_time_total_us": 29.088000000000648, "device_time_total_us": 3.6159999999999854, "self_device_time_total_us": 3.6159999999999854, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BUnaryFunctor >, std::array >(int, at::native::BUnaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.6159999999999854, "self_device_time_total_us": 3.6159999999999854, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 96.75199999999859, "self_cpu_time_total_us": 11.21600000000035, "device_time_total_us": 2.5279999999984284, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::_to_copy", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 85.53599999999824, "self_cpu_time_total_us": 23.519999999999527, "device_time_total_us": 2.5279999999984284, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 55.23199999999906, "self_cpu_time_total_us": 28.575999999999112, "device_time_total_us": 2.5279999999984284, "self_device_time_total_us": 2.5279999999984284, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.5279999999984284, "self_device_time_total_us": 2.5279999999984284, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Memset (Device)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.863999999997759, "self_device_time_total_us": 0.863999999997759, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 78306.49100000001, "self_cpu_time_total_us": 78306.49100000001, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 867.0880000000016, "self_cpu_time_total_us": 867.0880000000016, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 333.87200000000166, "self_cpu_time_total_us": 330.12800000000107, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::empty", "count": 12, "device_type": "DeviceType.CPU", "cpu_time_total_us": 53.36000000000058, "self_cpu_time_total_us": 53.36000000000058, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4568793088, "self_device_memory_usage_bytes": 4568793088, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cudaMemsetAsync", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 39.08799999999974, "self_cpu_time_total_us": 39.08799999999974, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 29.98400000000038, "self_cpu_time_total_us": 29.98400000000038, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 20.496000000001004, "self_cpu_time_total_us": 20.496000000001004, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 14.67200000000048, "self_cpu_time_total_us": 14.67200000000048, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "aten::resize_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 10.256000000000313, "self_cpu_time_total_us": 10.256000000000313, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813066240, "self_device_memory_usage_bytes": 813066240, "flops": 0, "input_shapes": "[[0], [], []]" }, { "key": "aten::reshape", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 9.263999999999214, "self_cpu_time_total_us": 6.143999999998414, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::empty_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.783999999999651, "self_cpu_time_total_us": 6.783999999999651, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.576000000000931, "self_cpu_time_total_us": 5.040000000002692, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 28672], [], [], [], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 4.87999999999829, "self_cpu_time_total_us": 4.87999999999829, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.7440000000005966, "self_cpu_time_total_us": 3.7440000000005966, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.6480000000010477, "self_cpu_time_total_us": 3.6480000000010477, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[28672, 336], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.1200000000008004, "self_cpu_time_total_us": 3.1200000000008004, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.040000000000873, "self_cpu_time_total_us": 2.2400000000016007, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.5359999999982392, "self_cpu_time_total_us": 1.5359999999982392, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 28672], [], [], []]" }, { "key": "aten::to", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.375999999999749, "self_cpu_time_total_us": 1.375999999999749, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.2639999999992142, "self_cpu_time_total_us": 1.2639999999992142, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.7999999999992724, "self_cpu_time_total_us": 0.7999999999992724, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672], []]" }, { "key": "[memory]", "count": 18, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -5381862400, "self_device_memory_usage_bytes": -5381862400, "flops": 0, "input_shapes": "[]" } ] }, { "name": "mlp_fc1", "quantizer": "vortex_recalculate", "module": { "class": "Nvfp4Linear", "in_features": 5376, "out_features": 28672, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 28672, 2688 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 28672, 336 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 5376 ], "input_is_contiguous": true, "input_stride": [ 5376, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 4.996801726520061e-06, "p50_s": 4.631991032510996e-06, "p90_s": 5.161581793799995e-06, "p95_s": 6.788797327317294e-06, "p99_s": 8.090569754131138e-06, "min_s": 4.447996616363525e-06, "max_s": 8.416012860834599e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 8.19360138848424e-06, "p50_s": 8.16801912151277e-06, "p90_s": 8.572766091674566e-06, "p95_s": 9.206385584548114e-06, "p99_s": 9.713281178846957e-06, "min_s": 7.503957021981478e-06, "max_s": 9.840005077421665e-06 }, "bias_cast": { "count": 10, "mean_s": 4.1727907955646515e-06, "p50_s": 4.160014213994145e-06, "p90_s": 4.3039792217314245e-06, "p95_s": 4.447990795597434e-06, "p99_s": 4.563200054690241e-06, "min_s": 3.936002030968666e-06, "max_s": 4.592002369463444e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.0016323688847478478, "p50_s": 0.0016255035006906837, "p90_s": 0.0016488469729665667, "p95_s": 0.0016816789662698283, "p99_s": 0.0017079445609124377, "min_s": 0.0016072000144049525, "max_s": 0.00171451095957309 }, "scale_compat_cast": { "count": 10, "mean_s": 1.3878295430913567e-05, "p50_s": 1.2615957530215383e-05, "p90_s": 1.8737633945420383e-05, "p95_s": 1.9536822219379242e-05, "p99_s": 2.0176172838546336e-05, "min_s": 1.1696014553308487e-05, "max_s": 2.0336010493338108e-05 }, "scale_to_device": { "count": 10, "mean_s": 1.1468795128166675e-05, "p50_s": 1.0991992894560099e-05, "p90_s": 1.2360006803646683e-05, "p95_s": 1.3619990204460917e-05, "p99_s": 1.4627976925112308e-05, "min_s": 1.0575982742011547e-05, "max_s": 1.4879973605275154e-05 }, "ck_quantize_nvfp4": { "count": 10, "mean_s": 0.0021427478175610306, "p50_s": 0.002115183015121147, "p90_s": 0.0022523318126332014, "p95_s": 0.002252965405932628, "p99_s": 0.0022534722805721687, "min_s": 0.0021083989995531738, "max_s": 0.0022535989992320538 }, "params_wrap": { "count": 10, "mean_s": 6.8128050770610574e-06, "p50_s": 6.791990017518401e-06, "p90_s": 7.233611540868878e-06, "p95_s": 7.24081473890692e-06, "p99_s": 7.246577297337353e-06, "min_s": 6.336020305752754e-06, "max_s": 7.2480179369449615e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.003836804605089128, "p50_s": 0.0038021259824745357, "p90_s": 0.003966951585607603, "p95_s": 0.004002778799622319, "p99_s": 0.0040314405708340925, "min_s": 0.003776589990593493, "max_s": 0.004038606013637036 }, "linear": { "count": 10, "mean_s": 0.034088603203417736, "p50_s": 0.034148512000683695, "p90_s": 0.034465921606170014, "p95_s": 0.03457616078958381, "p99_s": 0.03466435213631485, "min_s": 0.033573855995200574, "max_s": 0.034686399972997606 }, "slice_reshape": { "count": 10, "mean_s": 6.470386870205402e-06, "p50_s": 6.511982064694166e-06, "p90_s": 6.7823741119354965e-06, "p95_s": 6.919182487763464e-06, "p99_s": 7.0286291884258385e-06, "min_s": 6.127986125648022e-06, "max_s": 7.0559908635914326e-06 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 5376 ], "input_bytes": 406533120, "packed_qdata_bytes": 101670912, "block_scale_bytes": 12708864, "output_bytes": 2168176640, "separate_materialization_bytes": 2689089536, "epilogue_candidate": "BF16 split and SwiGLU input" }, "profiler_summary": { "device_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 4568796160, "negative_self_device_allocated_bytes": -4568796160, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/mlp_fc1-vortex_recalculate.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3680.063, "self_cpu_time_total_us": 1429.4720000000016, "device_time_total_us": 68406.907, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4337958912, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [28672, 5376], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2217.3909999999996, "self_cpu_time_total_us": 1829.5510000000013, "device_time_total_us": 68406.907, "self_device_time_total_us": 68403.547, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4337958912, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 2688], [28672, 2688], [], [], [37888, 336], [28672, 336], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 68403.547, "self_device_time_total_us": 68403.547, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "comfy_kitchen::quantize_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1975.7920000000013, "self_cpu_time_total_us": 491.7920000000049, "device_time_total_us": 4257.5679999999975, "self_device_time_total_us": 4212.2239999999965, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 230834176, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [], [], [], []]" }, { "key": "void comfy::(anonymous namespace)::quantize_nvfp4_kernel<__nv_bfloat16, __nv_fp4x2_e2m1, __nv_fp8_e4m3, true, true>(__nv_bfloat16 const*, float const*, __nv_fp4x2_e2m1*, __nv_fp8_e4m3*, unsigned long, unsigned long, unsigned long, unsigned long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4212.2239999999965, "self_device_time_total_us": 4212.2239999999965, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::partial_absmax_bf16_vec_kernel(uint4 const*, unsigned short const*, float*, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3363.5510000000068, "self_device_time_total_us": 3363.5510000000068, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1429.2639999999974, "self_cpu_time_total_us": 525.0559999999969, "device_time_total_us": 45.34400000000096, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 27492352, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 899.2639999999992, "self_cpu_time_total_us": 825.616, "device_time_total_us": 45.34400000000096, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 73.64799999999923, "self_cpu_time_total_us": 28.67199999999866, "device_time_total_us": 45.34400000000096, "self_device_time_total_us": 45.34400000000096, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 45.34400000000096, "self_device_time_total_us": 45.34400000000096, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1569.7269999999999, "self_cpu_time_total_us": 1278.831000000001, "device_time_total_us": 5.504000000000815, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 172.47999999999956, "self_cpu_time_total_us": 88.46399999999812, "device_time_total_us": 5.504000000000815, "self_device_time_total_us": 5.504000000000815, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "(anonymous namespace)::final_scale_warp_kernel(float const*, float*, long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.9679999999971187, "self_device_time_total_us": 3.9679999999971187, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 315.9359999999997, "self_cpu_time_total_us": 249.74399999999878, "device_time_total_us": 3.36000000000422, "self_device_time_total_us": 3.36000000000422, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.36000000000422, "self_device_time_total_us": 3.36000000000422, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 44.91199999999844, "self_cpu_time_total_us": 4.287999999998647, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 2.9760000000005675, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1554.5429999999997, "self_cpu_time_total_us": 25.4399999999996, "device_time_total_us": 2.5280000000002474, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.5280000000002474, "self_device_time_total_us": 2.5280000000002474, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 69911.369, "self_cpu_time_total_us": 69911.369, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1079.4230000000034, "self_cpu_time_total_us": 1079.4230000000034, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 929.1040000000003, "self_cpu_time_total_us": 929.1040000000003, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 407.0560000000005, "self_cpu_time_total_us": 400.9440000000013, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 118.41599999999926, "self_cpu_time_total_us": 118.41599999999926, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 2048, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::empty", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 37.47200000000157, "self_cpu_time_total_us": 37.47200000000157, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4568793088, "self_device_memory_usage_bytes": 4568793088, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 33.19999999999891, "self_cpu_time_total_us": 26.591999999998734, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 28672], [], [], [], []]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 21.375999999998385, "self_cpu_time_total_us": 21.375999999998385, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 21.023999999999432, "self_cpu_time_total_us": 21.023999999999432, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 15.935999999999694, "self_cpu_time_total_us": 15.935999999999694, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "aten::reshape", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 15.072000000000116, "self_cpu_time_total_us": 7.775999999998021, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 7.2960000000020955, "self_cpu_time_total_us": 7.2960000000020955, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.608000000000175, "self_cpu_time_total_us": 6.608000000000175, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 28672], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.495999999999185, "self_cpu_time_total_us": 4.31999999999789, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.1119999999991705, "self_cpu_time_total_us": 6.1119999999991705, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.992000000000189, "self_cpu_time_total_us": 2.992000000000189, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[28672, 336], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.9440000000013242, "self_cpu_time_total_us": 2.9440000000013242, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.176000000001295, "self_cpu_time_total_us": 2.176000000001295, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672], []]" }, { "key": "[memory]", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -4568795136, "self_device_memory_usage_bytes": -4568795136, "flops": 0, "input_shapes": "[]" } ] }, { "name": "mlp_fc1", "quantizer": "vortex_native", "module": { "class": "Nvfp4Linear", "in_features": 5376, "out_features": 28672, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 28672, 2688 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 28672, 336 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 5376 ], "input_is_contiguous": true, "input_stride": [ 5376, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 5.4367876145988704e-06, "p50_s": 4.743982572108507e-06, "p90_s": 7.131200982257723e-06, "p95_s": 7.973608444444833e-06, "p99_s": 8.647534414194525e-06, "min_s": 4.4959597289562225e-06, "max_s": 8.816015906631947e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 9.887997293844819e-06, "p50_s": 8.920003892853856e-06, "p90_s": 1.2116803554818033e-05, "p95_s": 1.4730411930941038e-05, "p99_s": 1.6821298631839455e-05, "min_s": 7.792026735842228e-06, "max_s": 1.7344020307064056e-05 }, "bias_cast": { "count": 10, "mean_s": 4.305597394704819e-06, "p50_s": 4.223984433338046e-06, "p90_s": 4.609598545357585e-06, "p95_s": 4.760813317261636e-06, "p99_s": 4.881785134784878e-06, "min_s": 4.047993570566177e-06, "max_s": 4.9120280891656876e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.0016478794103022664, "p50_s": 0.0016467270033899695, "p90_s": 0.0016853383334819226, "p95_s": 0.0016977731749648227, "p99_s": 0.001707721048151143, "min_s": 0.0016089430428110063, "max_s": 0.001710208016447723 }, "scale_compat_cast": { "count": 10, "mean_s": 0.00011893280898220838, "p50_s": 1.4992023352533579e-05, "p90_s": 0.0001227600208949294, "p95_s": 0.0005900040006963526, "p99_s": 0.0009637991845374928, "min_s": 1.1680007446557283e-05, "max_s": 0.0010572479804977775 }, "scale_to_device": { "count": 10, "mean_s": 1.5512004029005767e-05, "p50_s": 1.1624011676758528e-05, "p90_s": 2.5340763386338946e-05, "p95_s": 2.6550376787781715e-05, "p99_s": 2.7518067508935928e-05, "min_s": 1.0560033842921257e-05, "max_s": 2.775999018922448e-05 }, "vortex_quantize_nvfp4": { "count": 10, "mean_s": 0.00223893690854311, "p50_s": 0.002212615014286712, "p90_s": 0.002293387893587351, "p95_s": 0.0023610539501532912, "p99_s": 0.0024151867954060437, "min_s": 0.0021939509897492826, "max_s": 0.0024287200067192316 }, "params_wrap": { "count": 10, "mean_s": 9.187200339511037e-06, "p50_s": 6.959977326914668e-06, "p90_s": 9.856035467237227e-06, "p95_s": 1.964803668670354e-05, "p99_s": 2.748163766227663e-05, "min_s": 6.320013199001551e-06, "max_s": 2.944003790616989e-05 }, "activation_quantize": { "count": 10, "mean_s": 0.004063829098595306, "p50_s": 0.0039235510339494795, "p90_s": 0.00427848370745778, "p95_s": 0.004657296853838488, "p99_s": 0.004960347370943054, "min_s": 0.003862606012262404, "max_s": 0.005036110000219196 }, "linear": { "count": 10, "mean_s": 0.051964699605014174, "p50_s": 0.051965390506666154, "p90_s": 0.07021930979099125, "p95_s": 0.07034799542161636, "p99_s": 0.07045094392611645, "min_s": 0.03307627700269222, "max_s": 0.07047668105224147 }, "slice_reshape": { "count": 10, "mean_s": 7.37918890081346e-06, "p50_s": 6.959977326914668e-06, "p90_s": 8.595181861892343e-06, "p95_s": 9.113576379604637e-06, "p99_s": 9.528291993774474e-06, "min_s": 6.175949238240719e-06, "max_s": 9.631970897316933e-06 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 5376 ], "input_bytes": 406533120, "packed_qdata_bytes": 101670912, "block_scale_bytes": 12708864, "output_bytes": 2168176640, "separate_materialization_bytes": 2689089536, "epilogue_candidate": "BF16 split and SwiGLU input" }, "profiler_summary": { "device_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 4568796160, "negative_self_device_allocated_bytes": -4568796160, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/mlp_fc1-vortex_native.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 727.5519999999997, "self_cpu_time_total_us": 141.0880000000002, "device_time_total_us": 67877.84599999999, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4337958912, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], [28672, 5376], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 569.2799999999997, "self_cpu_time_total_us": 453.4079999999999, "device_time_total_us": 67877.84599999999, "self_device_time_total_us": 67873.97399999999, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4337958912, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 2688], [28672, 2688], [], [], [37888, 336], [28672, 336], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 67873.97399999999, "self_device_time_total_us": 67873.97399999999, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::quantize_nvfp4_bf16_kernel(unsigned short const*, float const*, unsigned char*, unsigned char*, long, long, long, long, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4261.194000000005, "self_device_time_total_us": 4261.194000000005, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::partial_absmax_bf16_vec_kernel(uint4 const*, unsigned short const*, float*, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3239.6230000000014, "self_device_time_total_us": 3239.6230000000014, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 62.19199999999978, "self_cpu_time_total_us": 15.568000000000666, "device_time_total_us": 45.792000000002645, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 27492352, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 41.90399999999954, "self_cpu_time_total_us": 10.67199999999957, "device_time_total_us": 45.792000000002645, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 31.23199999999997, "self_cpu_time_total_us": 12.672000000000025, "device_time_total_us": 45.792000000002645, "self_device_time_total_us": 45.792000000002645, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 45.792000000002645, "self_device_time_total_us": 45.792000000002645, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 273.9520000000002, "self_cpu_time_total_us": 34.40000000000077, "device_time_total_us": 5.472000000000662, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 140.5119999999995, "self_cpu_time_total_us": 56.6400000000001, "device_time_total_us": 5.472000000000662, "self_device_time_total_us": 5.472000000000662, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "(anonymous namespace)::final_scale_warp_kernel(float const*, float*, long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.9370000000012624, "self_device_time_total_us": 3.9370000000012624, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 42.14399999999978, "self_cpu_time_total_us": 24.1279999999997, "device_time_total_us": 3.87199999999757, "self_device_time_total_us": 3.87199999999757, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.87199999999757, "self_device_time_total_us": 3.87199999999757, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 41.69600000000014, "self_cpu_time_total_us": 3.423999999999978, "device_time_total_us": 2.9760000000010223, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.9760000000010223, "self_device_time_total_us": 2.9760000000010223, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 256.0639999999999, "self_cpu_time_total_us": 20.383999999999844, "device_time_total_us": 2.49599999999964, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.49599999999964, "self_device_time_total_us": 2.49599999999964, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 73904.807, "self_cpu_time_total_us": 73904.807, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1172.5439999999999, "self_cpu_time_total_us": 1172.5439999999999, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 197.6479999999999, "self_cpu_time_total_us": 197.6479999999999, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 99.03999999999996, "self_cpu_time_total_us": 99.03999999999996, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 2048, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::empty", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 34.95999999999913, "self_cpu_time_total_us": 34.95999999999913, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 4568793088, "self_device_memory_usage_bytes": 4568793088, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 31.695999999999913, "self_cpu_time_total_us": 31.695999999999913, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 17.18399999999974, "self_cpu_time_total_us": 12.495999999999185, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 28672], [], [], [], []]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 17.088000000000193, "self_cpu_time_total_us": 17.088000000000193, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 14.832000000000136, "self_cpu_time_total_us": 9.040000000000362, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.431999999999789, "self_cpu_time_total_us": 6.431999999999789, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 336], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.320000000000164, "self_cpu_time_total_us": 3.6159999999999854, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.791999999999774, "self_cpu_time_total_us": 5.791999999999774, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 4.688000000000557, "self_cpu_time_total_us": 4.688000000000557, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 28672], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.6480000000001382, "self_cpu_time_total_us": 1.8080000000004475, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.9440000000004147, "self_cpu_time_total_us": 2.9440000000004147, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.7040000000001783, "self_cpu_time_total_us": 2.7040000000001783, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.8399999999996908, "self_cpu_time_total_us": 1.8399999999996908, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 28672], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.3600000000001273, "self_cpu_time_total_us": 1.3600000000001273, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[28672, 336], []]" }, { "key": "[memory]", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -4568795136, "self_device_memory_usage_bytes": -4568795136, "flops": 0, "input_shapes": "[]" } ] }, { "name": "mlp_fc2", "quantizer": "comfy", "module": { "class": "Nvfp4Linear", "in_features": 14336, "out_features": 5376, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 5376, 7168 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 5376, 896 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 14336 ], "input_is_contiguous": true, "input_stride": [ 14336, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 5.179201252758503e-06, "p50_s": 4.592002369463444e-06, "p90_s": 5.366385448724029e-06, "p95_s": 7.843185449019069e-06, "p99_s": 9.824625449255109e-06, "min_s": 4.511966835707426e-06, "max_s": 1.0319985449314117e-05 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 8.53920355439186e-06, "p50_s": 8.416012860834599e-06, "p90_s": 8.804787648841739e-06, "p95_s": 9.618382318876683e-06, "p99_s": 1.0269258054904639e-05, "min_s": 7.935974281281233e-06, "max_s": 1.0431976988911629e-05 }, "bias_cast": { "count": 10, "mean_s": 4.128011642023921e-06, "p50_s": 4.056026227772236e-06, "p90_s": 4.284823080524802e-06, "p95_s": 4.486434045247733e-06, "p99_s": 4.647722817026079e-06, "min_s": 3.999972250312567e-06, "max_s": 4.688045009970665e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.01925712469383143, "p50_s": 0.019156786001985893, "p90_s": 0.019557130028260872, "p95_s": 0.019652889997814782, "p99_s": 0.01972949797345791, "min_s": 0.01911408902378753, "max_s": 0.019748649967368692 }, "linear": { "count": 10, "mean_s": 0.0551715788024012, "p50_s": 0.055235853971680626, "p90_s": 0.05537108918651938, "p95_s": 0.055614175589289515, "p99_s": 0.05580864471150562, "min_s": 0.054014783003367484, "max_s": 0.05585726199205965 }, "slice_reshape": { "count": 10, "mean_s": 6.69279252178967e-06, "p50_s": 6.6559878177940845e-06, "p90_s": 6.865593604743481e-06, "p95_s": 7.16079375706613e-06, "p99_s": 7.3969538789242504e-06, "min_s": 6.4159976318478584e-06, "max_s": 7.455993909388781e-06 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 14336 ], "input_bytes": 1084088320, "packed_qdata_bytes": 271122432, "block_scale_bytes": 33890304, "output_bytes": 406533120, "separate_materialization_bytes": 1795634176, "epilogue_candidate": "segmented BF16 residual gate/add" }, "profiler_summary": { "device_kernel_launches": 18, "unique_device_kernels": 9, "positive_self_device_allocated_bytes": 3591688192, "negative_self_device_allocated_bytes": -3591688192, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/mlp_fc2-comfy.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 684.2070000000003, "self_cpu_time_total_us": 126.79999999999927, "device_time_total_us": 110675.38, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], [5376, 14336], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 546.335, "self_cpu_time_total_us": 425.0710000000008, "device_time_total_us": 110675.38, "self_device_time_total_us": 110671.54, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 7168], [5376, 7168], [], [], [37888, 896], [5376, 896], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu_stream_k", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 110671.54000000001, "self_device_time_total_us": 110671.54000000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::abs", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1181.8869999999997, "self_cpu_time_total_us": 15.695999999999685, "device_time_total_us": 26738.947999999997, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2168176640, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336]]" }, { "key": "aten::abs", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1146.1910000000005, "self_cpu_time_total_us": 998.768, "device_time_total_us": 26738.947999999997, "self_device_time_total_us": 17820.685999999998, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2168176640, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], [0]]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::AbsFunctor, std::array >(int, at::native::AbsFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 17820.685999999998, "self_device_time_total_us": 17820.685999999998, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "comfy_kitchen::quantize_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 616.7360000000003, "self_cpu_time_total_us": 492.43200000000024, "device_time_total_us": 11358.48500000001, "self_device_time_total_us": 11129.75000000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 610140160, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], [], [], [], []]" }, { "key": "void comfy::(anonymous namespace)::quantize_nvfp4_kernel<__nv_bfloat16, __nv_fp4x2_e2m1, __nv_fp8_e4m3, true, true>(__nv_bfloat16 const*, float const*, __nv_fp4x2_e2m1*, __nv_fp8_e4m3*, unsigned long, unsigned long, unsigned long, unsigned long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 11129.75000000001, "self_device_time_total_us": 11129.75000000001, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 63.29500000000007, "self_cpu_time_total_us": 63.29500000000007, "device_time_total_us": 8918.261999999999, "self_device_time_total_us": 8918.261999999999, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::amax", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 528.1280000000006, "self_cpu_time_total_us": 350.1120000000001, "device_time_total_us": 8901.593000000004, "self_device_time_total_us": 8901.593000000004, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[37810, 14336], [], []]" }, { "key": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp >, unsigned int, c10::BFloat16, 4, 4> >(at::native::ReduceOp >, unsigned int, c10::BFloat16, 4, 4>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 8900.761000000006, "self_device_time_total_us": 8900.761000000006, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 69.60000000000036, "self_cpu_time_total_us": 18.736000000000786, "device_time_total_us": 228.73500000000058, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 67895296, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 46.7199999999998, "self_cpu_time_total_us": 14.831999999999425, "device_time_total_us": 228.73500000000058, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 31.888000000000375, "self_cpu_time_total_us": 17.135999999999967, "device_time_total_us": 228.73500000000058, "self_device_time_total_us": 228.73500000000058, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 228.73500000000058, "self_device_time_total_us": 228.73500000000058, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 52.159999999999854, "self_cpu_time_total_us": 29.040000000000873, "device_time_total_us": 3.8400000000074215, "self_device_time_total_us": 3.8400000000074215, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.8400000000074215, "self_device_time_total_us": 3.8400000000074215, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::div", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 67.35999999999967, "self_cpu_time_total_us": 45.2800000000002, "device_time_total_us": 3.679999999993015, "self_device_time_total_us": 3.679999999993015, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BUnaryFunctor >, std::array >(int, at::native::BUnaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.679999999993015, "self_device_time_total_us": 3.679999999993015, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 155.19999999999982, "self_cpu_time_total_us": 11.456000000000131, "device_time_total_us": 2.463999999999942, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::_to_copy", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 143.7439999999997, "self_cpu_time_total_us": 31.615999999999985, "device_time_total_us": 2.463999999999942, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 97.23199999999997, "self_cpu_time_total_us": 36.95999999999958, "device_time_total_us": 2.463999999999942, "self_device_time_total_us": 2.463999999999942, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.463999999999942, "self_device_time_total_us": 2.463999999999942, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Memset (Device)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.8319999999985157, "self_device_time_total_us": 0.8319999999985157, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 144756.801, "self_cpu_time_total_us": 144756.801, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 267.56799999999976, "self_cpu_time_total_us": 267.56799999999976, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaMemsetAsync", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 85.11999999999989, "self_cpu_time_total_us": 85.11999999999989, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty", "count": 12, "device_type": "DeviceType.CPU", "cpu_time_total_us": 52.351999999998895, "self_cpu_time_total_us": 52.351999999998895, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1423507456, "self_device_memory_usage_bytes": 1423507456, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::resize_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 39.648000000000366, "self_cpu_time_total_us": 39.648000000000366, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2168176640, "self_device_memory_usage_bytes": 2168176640, "flops": 0, "input_shapes": "[[0], [], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 24.720000000000255, "self_cpu_time_total_us": 24.720000000000255, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 19.071999999999207, "self_cpu_time_total_us": 19.071999999999207, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 15.32799999999952, "self_cpu_time_total_us": 9.135999999999058, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::empty_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 14.89599999999973, "self_cpu_time_total_us": 14.89599999999973, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 13.344000000000023, "self_cpu_time_total_us": 8.079999999999757, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 11.903999999999996, "self_cpu_time_total_us": 11.903999999999996, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 11.16800000000012, "self_cpu_time_total_us": 11.16800000000012, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896], []]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 11.072000000001026, "self_cpu_time_total_us": 7.792000000000371, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.431999999999789, "self_cpu_time_total_us": 3.5039999999999054, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.192000000000462, "self_cpu_time_total_us": 6.192000000000462, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.264000000000266, "self_cpu_time_total_us": 5.264000000000266, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], []]" }, { "key": "aten::to", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.488000000000426, "self_cpu_time_total_us": 3.488000000000426, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.4400000000005093, "self_cpu_time_total_us": 3.4400000000005093, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[5376, 896], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.280000000000655, "self_cpu_time_total_us": 3.280000000000655, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.9600000000000364, "self_cpu_time_total_us": 2.9600000000000364, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.9279999999998836, "self_cpu_time_total_us": 2.9279999999998836, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "[memory]", "count": 18, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -3591687168, "self_device_memory_usage_bytes": -3591687168, "flops": 0, "input_shapes": "[]" } ] }, { "name": "mlp_fc2", "quantizer": "vortex_recalculate", "module": { "class": "Nvfp4Linear", "in_features": 14336, "out_features": 5376, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 5376, 7168 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 5376, 896 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 14336 ], "input_is_contiguous": true, "input_stride": [ 14336, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 4.913605516776443e-06, "p50_s": 4.656030796468258e-06, "p90_s": 5.064037395641207e-06, "p95_s": 6.180009222589431e-06, "p99_s": 7.072786684148015e-06, "min_s": 4.4800108298659325e-06, "max_s": 7.295981049537659e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 8.223997429013253e-06, "p50_s": 8.096016244962811e-06, "p90_s": 8.585641626268624e-06, "p95_s": 9.06084314920008e-06, "p99_s": 9.441004367545247e-06, "min_s": 7.647962775081396e-06, "max_s": 9.536044672131538e-06 }, "bias_cast": { "count": 10, "mean_s": 4.2159983422607185e-06, "p50_s": 4.0959566831588745e-06, "p90_s": 4.289590287953615e-06, "p95_s": 4.800781607627867e-06, "p99_s": 5.209734663367272e-06, "min_s": 4.00003045797348e-06, "max_s": 5.311972927302122e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.0042167027073446665, "p50_s": 0.004208286991342902, "p90_s": 0.004253056499874219, "p95_s": 0.004259255257784389, "p99_s": 0.004264214264112525, "min_s": 0.00419579102890566, "max_s": 0.004265454015694559 }, "scale_compat_cast": { "count": 10, "mean_s": 1.2134405551478267e-05, "p50_s": 1.1976022506132722e-05, "p90_s": 1.271362998522818e-05, "p95_s": 1.2972814147360623e-05, "p99_s": 1.3180161477066575e-05, "min_s": 1.1487980373203754e-05, "max_s": 1.3231998309493065e-05 }, "scale_to_device": { "count": 10, "mean_s": 1.143038971349597e-05, "p50_s": 1.084798714146018e-05, "p90_s": 1.190401380881667e-05, "p95_s": 1.399199827574193e-05, "p99_s": 1.5662385849282145e-05, "min_s": 1.0607996955513954e-05, "max_s": 1.6079982742667198e-05 }, "ck_quantize_nvfp4": { "count": 10, "mean_s": 0.005730823596240953, "p50_s": 0.005727157986257225, "p90_s": 0.005760842794552445, "p95_s": 0.0057634924189187585, "p99_s": 0.00576561211841181, "min_s": 0.005701629968825728, "max_s": 0.005766142043285072 }, "params_wrap": { "count": 10, "mean_s": 6.742402911186219e-06, "p50_s": 6.688002031296492e-06, "p90_s": 6.943981861695647e-06, "p95_s": 7.159999222494661e-06, "p99_s": 7.332813111133874e-06, "min_s": 6.479967851191759e-06, "max_s": 7.376016583293676e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.01000769910751842, "p50_s": 0.010017916531069204, "p90_s": 0.010028161690570414, "p95_s": 0.010029766825027764, "p99_s": 0.010031050932593644, "min_s": 0.009965389035642147, "max_s": 0.010031371959485114 }, "linear": { "count": 10, "mean_s": 0.05473228610935621, "p50_s": 0.05463681402034126, "p90_s": 0.05536423642770387, "p95_s": 0.055373805234557955, "p99_s": 0.05538146028004121, "min_s": 0.05412913399050012, "max_s": 0.055383374041412026 }, "slice_reshape": { "count": 10, "mean_s": 6.51679583825171e-06, "p50_s": 6.4480118453502655e-06, "p90_s": 6.679981015622616e-06, "p95_s": 6.859982386231422e-06, "p99_s": 7.003983482718468e-06, "min_s": 6.336020305752754e-06, "max_s": 7.039983756840229e-06 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 14336 ], "input_bytes": 1084088320, "packed_qdata_bytes": 271122432, "block_scale_bytes": 33890304, "output_bytes": 406533120, "separate_materialization_bytes": 1795634176, "epilogue_candidate": "segmented BF16 residual gate/add" }, "profiler_summary": { "device_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 1423510528, "negative_self_device_allocated_bytes": -1423510528, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/mlp_fc2-vortex_recalculate.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 231.60000000000002, "self_cpu_time_total_us": 31.79200000000003, "device_time_total_us": 109667.01, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], [5376, 14336], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 192.38400000000001, "self_cpu_time_total_us": 139.2159999999999, "device_time_total_us": 109667.01, "self_device_time_total_us": 109663.04200000002, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 7168], [5376, 7168], [], [], [37888, 896], [5376, 896], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu_stream_k", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 109663.04200000002, "self_device_time_total_us": 109663.04200000002, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "comfy_kitchen::quantize_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 153.48799999999994, "self_cpu_time_total_us": 98.11199999999985, "device_time_total_us": 11601.821000000004, "self_device_time_total_us": 11372.093000000003, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 610140160, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], [], [], [], []]" }, { "key": "void comfy::(anonymous namespace)::quantize_nvfp4_kernel<__nv_bfloat16, __nv_fp4x2_e2m1, __nv_fp8_e4m3, true, true>(__nv_bfloat16 const*, float const*, __nv_fp4x2_e2m1*, __nv_fp8_e4m3*, unsigned long, unsigned long, unsigned long, unsigned long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 11372.093000000003, "self_device_time_total_us": 11372.093000000003, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::partial_absmax_bf16_vec_kernel(uint4 const*, unsigned short const*, float*, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 8416.318000000005, "self_device_time_total_us": 8416.318000000005, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 34.81599999999992, "self_cpu_time_total_us": 5.711999999999989, "device_time_total_us": 229.72800000000188, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 67895296, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 25.19999999999999, "self_cpu_time_total_us": 6.064000000000021, "device_time_total_us": 229.72800000000188, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 19.135999999999967, "self_cpu_time_total_us": 5.7439999999999145, "device_time_total_us": 229.72800000000188, "self_device_time_total_us": 229.72800000000188, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 229.72800000000188, "self_device_time_total_us": 229.72800000000188, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 67.92000000000002, "self_cpu_time_total_us": 12.048000000000059, "device_time_total_us": 5.5039999999926295, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 37.567999999999984, "self_cpu_time_total_us": 17.52000000000004, "device_time_total_us": 5.5039999999926295, "self_device_time_total_us": 5.5039999999926295, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "(anonymous namespace)::final_scale_warp_kernel(float const*, float*, long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 4.064000000002125, "self_device_time_total_us": 4.064000000002125, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 20.831999999999937, "self_cpu_time_total_us": 10.287999999999954, "device_time_total_us": 3.9679999999880238, "self_device_time_total_us": 3.9679999999880238, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.9679999999880238, "self_device_time_total_us": 3.9679999999880238, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 23.85600000000005, "self_cpu_time_total_us": 2.1440000000000623, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 2.9760000000005675, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 52.95999999999998, "self_cpu_time_total_us": 6.751999999999953, "device_time_total_us": 2.527999999992062, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.527999999992062, "self_device_time_total_us": 2.527999999992062, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 129138.23, "self_cpu_time_total_us": 129138.23, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 87.95199999999986, "self_cpu_time_total_us": 87.95199999999986, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 64.624, "self_cpu_time_total_us": 64.624, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 20.767999999999972, "self_cpu_time_total_us": 20.767999999999972, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1423507456, "self_device_memory_usage_bytes": 1423507456, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 18.303999999999974, "self_cpu_time_total_us": 18.303999999999974, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 2048, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 14.960000000000036, "self_cpu_time_total_us": 14.960000000000036, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 7.696000000000012, "self_cpu_time_total_us": 5.744000000000042, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], []]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 7.423999999999978, "self_cpu_time_total_us": 6.447999999999979, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], [], []]" }, { "key": "aten::reshape", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.760000000000048, "self_cpu_time_total_us": 4.048000000000002, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.4720000000000937, "self_cpu_time_total_us": 3.4720000000000937, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.5120000000000573, "self_cpu_time_total_us": 1.8400000000001455, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.2880000000000678, "self_cpu_time_total_us": 2.2880000000000678, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.9519999999999698, "self_cpu_time_total_us": 1.9519999999999698, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.712000000000046, "self_cpu_time_total_us": 1.712000000000046, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.1200000000001182, "self_cpu_time_total_us": 1.1200000000001182, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[5376, 896], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.9759999999999991, "self_cpu_time_total_us": 0.9759999999999991, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.9279999999999973, "self_cpu_time_total_us": 0.9279999999999973, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.6719999999999118, "self_cpu_time_total_us": 0.6719999999999118, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "[memory]", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -1423509504, "self_device_memory_usage_bytes": -1423509504, "flops": 0, "input_shapes": "[]" } ] }, { "name": "mlp_fc2", "quantizer": "vortex_native", "module": { "class": "Nvfp4Linear", "in_features": 14336, "out_features": 5376, "output_dtype": "torch.bfloat16", "full_precision_matrix_mult": false, "weight_dtype": "torch.uint8", "weight_shape": [ 5376, 7168 ], "weight_scale_dtype": "torch.float8_e4m3fn", "weight_scale_shape": [ 5376, 896 ], "weight_scale_2_dtype": "torch.float32", "weight_scale_2_shape": [], "bias_dtype": null, "bias_shape": null, "pre_quant_scale": false, "input_dtype": "torch.bfloat16", "input_shape": [ 37810, 14336 ], "input_is_contiguous": true, "input_stride": [ 14336, 1 ] }, "stage_timings": { "flatten_contiguous": { "count": 10, "mean_s": 4.836788866668939e-06, "p50_s": 4.592002369463444e-06, "p90_s": 5.012779729440807e-06, "p95_s": 5.97038597334176e-06, "p99_s": 6.736470968462527e-06, "min_s": 4.415982402861118e-06, "max_s": 6.927992217242718e-06 }, "pre_quant_scale": { "count": 10, "mean_s": 0.0, "p50_s": 0.0, "p90_s": 0.0, "p95_s": 0.0, "p99_s": 0.0, "min_s": 0.0, "max_s": 0.0 }, "packed_weight_wrapper": { "count": 10, "mean_s": 8.192012319341303e-06, "p50_s": 8.055998478084803e-06, "p90_s": 8.614402031525969e-06, "p95_s": 9.075223351828752e-06, "p99_s": 9.443880408070982e-06, "min_s": 7.823982741683722e-06, "max_s": 9.536044672131538e-06 }, "bias_cast": { "count": 10, "mean_s": 4.124792758375406e-06, "p50_s": 4.0800077840685844e-06, "p90_s": 4.228804027661681e-06, "p95_s": 4.322393215261399e-06, "p99_s": 4.397264565341175e-06, "min_s": 3.999972250312567e-06, "max_s": 4.415982402861118e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.004220321995671838, "p50_s": 0.004209934995742515, "p90_s": 0.004265264485729858, "p95_s": 0.004276431250036694, "p99_s": 0.004285364661482162, "min_s": 0.004186158999800682, "max_s": 0.00428759801434353 }, "scale_compat_cast": { "count": 10, "mean_s": 1.2371200136840344e-05, "p50_s": 1.1800002539530396e-05, "p90_s": 1.2766424333676694e-05, "p95_s": 1.5135228750295932e-05, "p99_s": 1.703027228359133e-05, "min_s": 1.1391995940357447e-05, "max_s": 1.7504033166915178e-05 }, "scale_to_device": { "count": 10, "mean_s": 1.1740811169147492e-05, "p50_s": 1.096801133826375e-05, "p90_s": 1.2745597632601854e-05, "p95_s": 1.530880981590598e-05, "p99_s": 1.7359379562549293e-05, "min_s": 1.057604094967246e-05, "max_s": 1.787202199921012e-05 }, "vortex_quantize_nvfp4": { "count": 10, "mean_s": 0.005958578817080706, "p50_s": 0.005956517998129129, "p90_s": 0.00597937082638964, "p95_s": 0.0059890044096391655, "p99_s": 0.0059967112762387845, "min_s": 0.005931598017923534, "max_s": 0.005998637992888689 }, "params_wrap": { "count": 10, "mean_s": 6.75680348649621e-06, "p50_s": 6.704009138047695e-06, "p90_s": 7.038423791527748e-06, "p95_s": 7.175205973908305e-06, "p99_s": 7.284631719812751e-06, "min_s": 6.52798917144537e-06, "max_s": 7.311988156288862e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.010239630402065814, "p50_s": 0.010224309022305533, "p90_s": 0.010317761672195048, "p95_s": 0.010330742830410598, "p99_s": 0.010341127756983043, "min_s": 0.010191901004873216, "max_s": 0.010343723988626152 }, "linear": { "count": 10, "mean_s": 0.054607604414923114, "p50_s": 0.054261310026049614, "p90_s": 0.055418743623886255, "p95_s": 0.05567892282269895, "p99_s": 0.05588706618174911, "min_s": 0.05413044604938477, "max_s": 0.055939102021511644 }, "slice_reshape": { "count": 10, "mean_s": 6.523198680952191e-06, "p50_s": 6.432004738599062e-06, "p90_s": 6.795220542699098e-06, "p95_s": 7.061607902869582e-06, "p99_s": 7.27471779100597e-06, "min_s": 6.304006092250347e-06, "max_s": 7.327995263040066e-06 } }, "reference_diff": { "max": 0.0, "mean": 0.0 }, "intermediate_traffic": { "logical_rows": 37810, "padded_activation_shape": [ 37824, 14336 ], "input_bytes": 1084088320, "packed_qdata_bytes": 271122432, "block_scale_bytes": 33890304, "output_bytes": 406533120, "separate_materialization_bytes": 1795634176, "epilogue_candidate": "segmented BF16 residual gate/add" }, "profiler_summary": { "device_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 1423510528, "negative_self_device_allocated_bytes": -1423510528, "trace_path": "/output/h3-blackwell-runtime/benchmarks/traces/gb10-nvfp4-component2/mlp_fc2-vortex_native.json", "note": "Profiler memory values are aggregate event attribution, not peak-memory measurements." }, "profiler_top_events": [ { "key": "aten::linear", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 682.9750000000001, "self_cpu_time_total_us": 108.75199999999995, "device_time_total_us": 109631.81999999998, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], [5376, 14336], []]" }, { "key": "comfy_kitchen::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 562.2230000000002, "self_cpu_time_total_us": 466.54300000000023, "device_time_total_us": 109631.81999999998, "self_device_time_total_us": 109628.044, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 813367296, "self_device_memory_usage_bytes": -1024, "flops": 0, "input_shapes": "[[37824, 7168], [5376, 7168], [], [], [37888, 896], [5376, 896], [], [], []]" }, { "key": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu_stream_k", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 109628.044, "self_device_time_total_us": 109628.044, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::quantize_nvfp4_bf16_kernel(unsigned short const*, float const*, unsigned char*, unsigned char*, long, long, long, long, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 12120.637000000006, "self_device_time_total_us": 12120.637000000006, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "(anonymous namespace)::partial_absmax_bf16_vec_kernel(uint4 const*, unsigned short const*, float*, long, long)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 8392.425000000003, "self_device_time_total_us": 8392.425000000003, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 52.2879999999999, "self_cpu_time_total_us": 10.62399999999991, "device_time_total_us": 252.70399999999972, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 67895296, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 37.039999999999964, "self_cpu_time_total_us": 9.631999999999948, "device_time_total_us": 252.70399999999972, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896]]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 27.408000000000015, "self_cpu_time_total_us": 12.912000000000035, "device_time_total_us": 252.70399999999972, "self_device_time_total_us": 252.70399999999972, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor, std::array >(int, at::native::FillFunctor, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 252.70399999999972, "self_device_time_total_us": 252.70399999999972, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 216.4319999999998, "self_cpu_time_total_us": 27.855999999999938, "device_time_total_us": 5.471999999992477, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 136.63999999999993, "self_cpu_time_total_us": 102.27200000000022, "device_time_total_us": 5.471999999992477, "self_device_time_total_us": 5.471999999992477, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], []]" }, { "key": "(anonymous namespace)::final_scale_warp_kernel(float const*, float*, long, float)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.936000000004242, "self_device_time_total_us": 3.936000000004242, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 44.59199999999976, "self_cpu_time_total_us": 26.0799999999997, "device_time_total_us": 3.775999999992564, "self_device_time_total_us": 3.775999999992564, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 1024, "flops": 2, "input_shapes": "[[], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor >, std::array >(int, at::native::BinaryFunctor >, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 3.775999999992564, "self_device_time_total_us": 3.775999999992564, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 34.94400000000019, "self_cpu_time_total_us": 2.496000000000322, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.9760000000005675, "self_device_time_total_us": 2.9760000000005675, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 195.51999999999978, "self_cpu_time_total_us": 11.53599999999986, "device_time_total_us": 2.495999999991909, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "void at::native::vectorized_elementwise_kernel<4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array >(int, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda(float)#1}, std::array)", "count": 2, "device_type": "DeviceType.CUDA", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 2.495999999991909, "self_device_time_total_us": 2.495999999991909, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaDeviceSynchronize", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 129225.382, "self_cpu_time_total_us": 129225.382, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "cudaLaunchKernel", "count": 14, "device_type": "DeviceType.CPU", "cpu_time_total_us": 142.22399999999948, "self_cpu_time_total_us": 142.22399999999948, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "Activity Buffer Request", "count": 1, "device_type": "DeviceType.CPU", "cpu_time_total_us": 62.70400000000001, "self_cpu_time_total_us": 62.70400000000001, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 51.93599999999992, "self_cpu_time_total_us": 51.93599999999992, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 2048, "self_device_memory_usage_bytes": 2048, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::empty", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 33.408000000000015, "self_cpu_time_total_us": 33.408000000000015, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 1423507456, "self_device_memory_usage_bytes": 1423507456, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cudaGetSymbolAddress", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 16.368000000000166, "self_cpu_time_total_us": 16.368000000000166, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 14.784000000000049, "self_cpu_time_total_us": 8.656000000000205, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 13.199999999999818, "self_cpu_time_total_us": 13.199999999999818, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[]" }, { "key": "aten::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 12.0, "self_cpu_time_total_us": 9.200000000000273, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.127999999999844, "self_cpu_time_total_us": 6.127999999999844, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 14336], []]" }, { "key": "aten::view", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.296000000000049, "self_cpu_time_total_us": 5.296000000000049, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37888, 896], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.616000000000099, "self_cpu_time_total_us": 2.240000000000123, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::as_strided", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.799999999999727, "self_cpu_time_total_us": 2.799999999999727, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37824, 5376], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.5920000000000982, "self_cpu_time_total_us": 1.7440000000001419, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.7440000000001419, "self_cpu_time_total_us": 1.7440000000001419, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[5376, 896], []]" }, { "key": "aten::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.456000000000131, "self_cpu_time_total_us": 1.456000000000131, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.3759999999999764, "self_cpu_time_total_us": 1.3759999999999764, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.8479999999999563, "self_cpu_time_total_us": 0.8479999999999563, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 0, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[37810, 5376], []]" }, { "key": "[memory]", "count": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.0, "self_cpu_time_total_us": 0.0, "device_time_total_us": 0.0, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": -1423509504, "self_device_memory_usage_bytes": -1423509504, "flops": 0, "input_shapes": "[]" } ] } ] }