{ "model_path": "/models/minimax_h3_fl2va_pruned_nvfp4.safetensors", "block_index": 24, "attention": "sage2", "warmup": 3, "iterations": 10, "profiler_iterations": 2, "quantizers": [ "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": "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": 6.052799290046096e-06, "p50_s": 5.256006261333823e-06, "p90_s": 9.011215297505258e-06, "p95_s": 9.889612556435166e-06, "p99_s": 1.0592330363579095e-05, "min_s": 4.047993570566177e-06, "max_s": 1.0768009815365076e-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": 1.6932794824242593e-05, "p50_s": 1.1727999662980437e-05, "p90_s": 3.8022396620363e-05, "p95_s": 3.869918291456997e-05, "p99_s": 3.924061194993555e-05, "min_s": 7.951981388032436e-06, "max_s": 3.937596920877695e-05 }, "bias_cast": { "count": 10, "mean_s": 3.7984049413353204e-06, "p50_s": 3.816006937995553e-06, "p90_s": 3.96320829167962e-06, "p95_s": 4.013604484498501e-06, "p99_s": 4.053921438753605e-06, "min_s": 3.5040429793298244e-06, "max_s": 4.064000677317381e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.0024285033985506745, "p50_s": 0.0025771994842216372, "p90_s": 0.002898441592697054, "p95_s": 0.0030190847872290758, "p99_s": 0.0031155993428546936, "min_s": 0.0016482880455441773, "max_s": 0.003139727981761098 }, "scale_compat_cast": { "count": 10, "mean_s": 2.7616001898422837e-05, "p50_s": 2.454398781992495e-05, "p90_s": 5.4996821563690905e-05, "p95_s": 5.516241653822362e-05, "p99_s": 5.5294892517849804e-05, "min_s": 1.02399499155581e-05, "max_s": 5.532801151275635e-05 }, "scale_to_device": { "count": 10, "mean_s": 1.5809608157724142e-05, "p50_s": 1.5712023014202714e-05, "p90_s": 2.2017589071765545e-05, "p95_s": 2.2168803843669595e-05, "p99_s": 2.2289775661192834e-05, "min_s": 9.231967851519585e-06, "max_s": 2.2320018615573645e-05 }, "vortex_quantize_nvfp4": { "count": 10, "mean_s": 0.0028375787020195275, "p50_s": 0.002920447994256392, "p90_s": 0.0030217359191738067, "p95_s": 0.003200979460962116, "p99_s": 0.0033443742943927647, "min_s": 0.002199806971475482, "max_s": 0.0033802230027504265 }, "params_wrap": { "count": 10, "mean_s": 7.174396887421608e-06, "p50_s": 6.7280197981745005e-06, "p90_s": 8.566409815102815e-06, "p95_s": 9.24319610930979e-06, "p99_s": 9.784625144675375e-06, "min_s": 6.127986125648022e-06, "max_s": 9.91998240351677e-06 }, "activation_quantize": { "count": 10, "mean_s": 0.005351344519294798, "p50_s": 0.0055619110353291035, "p90_s": 0.006045710993930697, "p95_s": 0.006317583000054583, "p99_s": 0.006535080604953692, "min_s": 0.003900239011272788, "max_s": 0.0065894550061784685 }, "linear": { "count": 10, "mean_s": 0.18048321470851078, "p50_s": 0.08908659202279523, "p90_s": 0.3484924691321794, "p95_s": 0.38921985907072654, "p99_s": 0.4218017710215645, "min_s": 0.02535125898430124, "max_s": 0.42994724900927395 }, "slice_reshape": { "count": 10, "mean_s": 1.6507197869941592e-05, "p50_s": 7.480004569515586e-06, "p90_s": 2.6808015536516882e-05, "p95_s": 5.6436000159010215e-05, "p99_s": 8.0138387857005e-05, "min_s": 5.759997293353081e-06, "max_s": 8.606398478150368e-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": 12730368, "output_bytes": 1626132480, "separate_materialization_bytes": 2147066880, "epilogue_candidate": "QKV views/layout and Q/K normalization/RoPE" }, "profiler_summary": { "device_kernel_launches": 0, "runtime_kernel_launches": 0, "unique_device_kernels": 0, "positive_self_device_allocated_bytes": 3482274816, "negative_self_device_allocated_bytes": -3482274816, "profiled_iterations": 2, "trace_path": null, "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": 3133.4399999999987, "self_cpu_time_total_us": 797.7919999999976, "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::scaled_mm_nvfp4", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2144.351999999999, "self_cpu_time_total_us": 1963.8880000000026, "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::_to_copy", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1378.5920000000042, "self_cpu_time_total_us": 874.9120000000039, "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": 2048, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], [], [], []]" }, { "key": "aten::to", "count": 6, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1296.912000000002, "self_cpu_time_total_us": 19.023999999999432, "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": 1024, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1074.0629999999983, "self_cpu_time_total_us": 364.5109999999986, "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": 25460736, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 919.5669999999991, "self_cpu_time_total_us": 911.1669999999995, "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::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 691.119999999999, "self_cpu_time_total_us": 636.5280000000021, "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::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 343.82400000000416, "self_cpu_time_total_us": 343.82400000000416, "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::slice", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 191.2960000000021, "self_cpu_time_total_us": 184.49600000000282, "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::copy_", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 159.85599999999613, "self_cpu_time_total_us": 159.85599999999613, "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::to", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 108.78399999999965, "self_cpu_time_total_us": 8.079999999998108, "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::mul", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 105.44000000000233, "self_cpu_time_total_us": 105.44000000000233, "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": 10, "device_type": "DeviceType.CPU", "cpu_time_total_us": 101.19999999999345, "self_cpu_time_total_us": 101.19999999999345, "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": 3482271744, "self_device_memory_usage_bytes": 3482271744, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "aten::fill_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 54.591999999996915, "self_cpu_time_total_us": 54.591999999996915, "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": "[[37888, 336], []]" }, { "key": "aten::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 15.711999999999534, "self_cpu_time_total_us": 8.272000000000844, "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": 10.11200000000099, "self_cpu_time_total_us": 10.11200000000099, "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": 8.399999999999636, "self_cpu_time_total_us": 8.399999999999636, "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": 8.351999999998952, "self_cpu_time_total_us": 5.328000000001339, "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": 7.43999999999869, "self_cpu_time_total_us": 7.43999999999869, "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": 6.799999999999272, "self_cpu_time_total_us": 6.799999999999272, "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::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 5.824000000000524, "self_cpu_time_total_us": 5.824000000000524, "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": 3.9200000000018917, "self_cpu_time_total_us": 3.9200000000018917, "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": 3.0239999999976135, "self_cpu_time_total_us": 3.0239999999976135, "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": -3482273792, "self_device_memory_usage_bytes": -3482273792, "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": 5.295989103615284e-06, "p50_s": 4.7759676817804575e-06, "p90_s": 7.12319160811603e-06, "p95_s": 7.425594958476721e-06, "p99_s": 7.667517638765274e-06, "min_s": 4.368019290268421e-06, "max_s": 7.727998308837414e-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": 9.599997429177166e-06, "p50_s": 8.68001370690763e-06, "p90_s": 1.0455993469804522e-05, "p95_s": 1.3947999104857437e-05, "p99_s": 1.674160361289978e-05, "min_s": 7.839989848434925e-06, "max_s": 1.7440004739910364e-05 }, "bias_cast": { "count": 10, "mean_s": 4.123203689232469e-06, "p50_s": 4.064000677317381e-06, "p90_s": 4.3728097807616e-06, "p95_s": 4.466398968361318e-06, "p99_s": 4.5412703184410926e-06, "min_s": 3.9199949242174625e-06, "max_s": 4.559988155961037e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.0021523084957152604, "p50_s": 0.0021477669943124056, "p90_s": 0.0021769711805973203, "p95_s": 0.0021807655895827336, "p99_s": 0.0021838011167710646, "min_s": 0.0021312159951776266, "max_s": 0.0021845599985681474 }, "scale_compat_cast": { "count": 10, "mean_s": 0.0002433982852380723, "p50_s": 1.305597834289074e-05, "p90_s": 0.0010138942569028586, "p95_s": 0.0011852826195536178, "p99_s": 0.0013223933096742258, "min_s": 1.1071970220655203e-05, "max_s": 0.0013566709822043777 }, "scale_to_device": { "count": 10, "mean_s": 1.2624007649719714e-05, "p50_s": 1.104001421481371e-05, "p90_s": 1.4812796143814918e-05, "p95_s": 2.0126372692175197e-05, "p99_s": 2.437723393086344e-05, "min_s": 1.0512012522667646e-05, "max_s": 2.5439949240535498e-05 }, "vortex_quantize_nvfp4": { "count": 10, "mean_s": 0.003013088909210637, "p50_s": 0.002953759511001408, "p90_s": 0.003206658235285431, "p95_s": 0.0032714726112317294, "p99_s": 0.003323324111988768, "min_s": 0.00292299099965021, "max_s": 0.0033362869871780276 }, "params_wrap": { "count": 10, "mean_s": 8.95520788617432e-06, "p50_s": 6.960006430745125e-06, "p90_s": 9.78240277618169e-06, "p95_s": 1.8379196990281324e-05, "p99_s": 2.525663236156106e-05, "min_s": 6.27199187874794e-06, "max_s": 2.697599120438099e-05 }, "activation_quantize": { "count": 10, "mean_s": 0.0054626068915240465, "p50_s": 0.005206414993153885, "p90_s": 0.006140521290944889, "p95_s": 0.006342604130622931, "p99_s": 0.0065042704023653645, "min_s": 0.005111566977575421, "max_s": 0.0065446869703009725 }, "linear": { "count": 10, "mean_s": 0.008279907511314378, "p50_s": 0.008283590519567952, "p90_s": 0.008627811732003466, "p95_s": 0.008640072858543135, "p99_s": 0.008649881759774871, "min_s": 0.007697679044213146, "max_s": 0.008652333985082805 }, "slice_reshape": { "count": 10, "mean_s": 6.809615297242999e-06, "p50_s": 6.5600324887782335e-06, "p90_s": 7.382442709058522e-06, "p95_s": 8.275220170617101e-06, "p99_s": 8.989442139863968e-06, "min_s": 6.127986125648022e-06, "max_s": 9.167997632175684e-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": 16973824, "output_bytes": 406533120, "separate_materialization_bytes": 1101112320, "epilogue_candidate": "segmented BF16 residual gate/add" }, "profiler_summary": { "device_kernel_launches": 16, "runtime_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 1118440448, "negative_self_device_allocated_bytes": -1118440448, "profiled_iterations": 2, "trace_path": null, "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": 308.0639999999994, "self_cpu_time_total_us": 54.4159999999988, "device_time_total_us": 15811.336000000001, "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": 246.5920000000001, "self_cpu_time_total_us": 196.46399999999994, "device_time_total_us": 15811.336000000001, "self_device_time_total_us": 15806.536, "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": 15806.536, "self_device_time_total_us": 15806.536, "cpu_memory_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": 6025.104000000001, "self_device_time_total_us": 6025.104000000001, "cpu_memory_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": 4324.332, "self_device_time_total_us": 4324.332, "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": "aten::zeros", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 31.744000000000142, "self_cpu_time_total_us": 6.144000000000233, "device_time_total_us": 82.94400000000041, "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": 20.783999999999878, "self_cpu_time_total_us": 6.479999999999791, "device_time_total_us": 82.94400000000041, "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": 14.304000000000087, "self_cpu_time_total_us": 5.824000000000069, "device_time_total_us": 82.94400000000041, "self_device_time_total_us": 82.94400000000041, "cpu_memory_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": 82.94400000000041, "self_device_time_total_us": 82.94400000000041, "cpu_memory_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": 79.83999999999992, "self_cpu_time_total_us": 16.62400000000025, "device_time_total_us": 5.824000000001888, "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": 43.24800000000005, "self_cpu_time_total_us": 21.07199999999989, "device_time_total_us": 5.824000000001888, "self_device_time_total_us": 5.824000000001888, "cpu_memory_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": 21.808000000000447, "self_cpu_time_total_us": 11.903999999999996, "device_time_total_us": 4.800000000002001, "self_device_time_total_us": 4.800000000002001, "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": 4.800000000002001, "self_device_time_total_us": 4.800000000002001, "cpu_memory_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": 4.032000000000153, "self_device_time_total_us": 4.032000000000153, "cpu_memory_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": 27.50400000000036, "self_cpu_time_total_us": 2.4000000000005457, "device_time_total_us": 3.2960000000002765, "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": 3.2960000000002765, "self_device_time_total_us": 3.2960000000002765, "cpu_memory_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": 63.4879999999998, "self_cpu_time_total_us": 8.751999999999697, "device_time_total_us": 2.5280000000016116, "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": 2.5280000000016116, "self_device_time_total_us": 2.5280000000016116, "cpu_memory_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": 24126.972, "self_cpu_time_total_us": 24126.972, "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": 1180.512, "self_cpu_time_total_us": 1180.512, "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": 77.74400000000082, "self_cpu_time_total_us": 77.74400000000082, "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": 21.792000000000144, "self_cpu_time_total_us": 21.792000000000144, "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": "aten::empty_strided", "count": 4, "device_type": "DeviceType.CPU", "cpu_time_total_us": 19.96799999999962, "self_cpu_time_total_us": 19.96799999999962, "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::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 8.47999999999982, "self_cpu_time_total_us": 6.687999999999647, "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": 8.111999999999625, "self_cpu_time_total_us": 8.111999999999625, "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": 7.056000000000495, "self_cpu_time_total_us": 5.296000000000731, "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": 4.096000000000004, "self_cpu_time_total_us": 2.800000000000182, "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": 2.8960000000001855, "self_cpu_time_total_us": 2.8960000000001855, "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": 2.7679999999995744, "self_cpu_time_total_us": 1.9679999999998472, "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": 2.127999999999929, "self_cpu_time_total_us": 2.127999999999929, "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::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.7920000000001721, "self_cpu_time_total_us": 1.7920000000001721, "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": 1.7599999999997635, "self_cpu_time_total_us": 1.7599999999997635, "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": 1.7280000000000655, "self_cpu_time_total_us": 1.7280000000000655, "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": 1.2959999999998217, "self_cpu_time_total_us": 1.2959999999998217, "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::alias", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 1.1840000000001965, "self_cpu_time_total_us": 1.1840000000001965, "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": 0.7999999999997272, "self_cpu_time_total_us": 0.7999999999997272, "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": "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": 5.855993367731571e-06, "p50_s": 5.0559756346046925e-06, "p90_s": 7.955211913213134e-06, "p95_s": 8.18558328319341e-06, "p99_s": 8.36988037917763e-06, "min_s": 4.496017936617136e-06, "max_s": 8.415954653173685e-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": 7.972479797899724e-05, "p50_s": 9.687995770946145e-06, "p90_s": 8.677759324200428e-05, "p95_s": 0.000390660794801078, "p99_s": 0.0006337673560483382, "min_s": 8.128001354634762e-06, "max_s": 0.000694543996360153 }, "bias_cast": { "count": 10, "mean_s": 4.559999797493219e-06, "p50_s": 4.111992893740535e-06, "p90_s": 4.9424183089286075e-06, "p95_s": 6.303194095380601e-06, "p99_s": 7.391814724542201e-06, "min_s": 3.968016244471073e-06, "max_s": 7.6639698818326e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.0016238380863796919, "p50_s": 0.0016251999768428504, "p90_s": 0.0016435487137641757, "p95_s": 0.001646701854770072, "p99_s": 0.0016492243675747887, "min_s": 0.001595902955159545, "max_s": 0.0016498549957759678 }, "scale_compat_cast": { "count": 10, "mean_s": 0.0022213820077013223, "p50_s": 1.616799272596836e-05, "p90_s": 0.010254466830519959, "p95_s": 0.01038646443921607, "p99_s": 0.010492062526172958, "min_s": 1.166400033980608e-05, "max_s": 0.01051846204791218 }, "scale_to_device": { "count": 10, "mean_s": 1.7052795737981796e-05, "p50_s": 1.2168020475655794e-05, "p90_s": 2.627038047648966e-05, "p95_s": 3.2023168751038596e-05, "p99_s": 3.662539937067777e-05, "min_s": 1.047999830916524e-05, "max_s": 3.777595702558756e-05 }, "vortex_quantize_nvfp4": { "count": 10, "mean_s": 0.002223322878126055, "p50_s": 0.0022234079951886088, "p90_s": 0.002246807888150215, "p95_s": 0.0022486434201709924, "p99_s": 0.0022501118457876144, "min_s": 0.002194703964050859, "max_s": 0.00225047895219177 }, "params_wrap": { "count": 10, "mean_s": 1.3769592624157666e-05, "p50_s": 7.775990525260568e-06, "p90_s": 2.7236773166805506e-05, "p95_s": 2.7474400121718644e-05, "p99_s": 2.7664501685649154e-05, "min_s": 6.383983418345451e-06, "max_s": 2.7712027076631784e-05 }, "activation_quantize": { "count": 10, "mean_s": 0.006139096606057137, "p50_s": 0.003941255999961868, "p90_s": 0.014208465768024326, "p95_s": 0.014347231370629742, "p99_s": 0.014458243852714076, "min_s": 0.0038519350346177816, "max_s": 0.01448599697323516 }, "linear": { "count": 10, "mean_s": 0.05182686430052854, "p50_s": 0.05186239897739142, "p90_s": 0.06942712950403802, "p95_s": 0.06958699877432081, "p99_s": 0.06971489419054706, "min_s": 0.033370858000125736, "max_s": 0.06974686804460362 }, "slice_reshape": { "count": 10, "mean_s": 7.93120707385242e-06, "p50_s": 6.83200778439641e-06, "p90_s": 1.0513601591810583e-05, "p95_s": 1.0592784383334219e-05, "p99_s": 1.0656130616553127e-05, "min_s": 6.176007445901632e-06, "max_s": 1.0671967174857855e-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": 12730368, "output_bytes": 2168176640, "separate_materialization_bytes": 2689111040, "epilogue_candidate": "BF16 split and SwiGLU input" }, "profiler_summary": { "device_kernel_launches": 16, "runtime_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 4566764544, "negative_self_device_allocated_bytes": -4566764544, "profiled_iterations": 2, "trace_path": null, "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": 473.4720000000002, "self_cpu_time_total_us": 84.49600000000032, "device_time_total_us": 66153.844, "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": 378.624, "self_cpu_time_total_us": 306.89600000000087, "device_time_total_us": 66153.844, "self_device_time_total_us": 66149.84399999998, "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": 66149.84399999998, "self_device_time_total_us": 66149.84399999998, "cpu_memory_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": 4390.723000000004, "self_device_time_total_us": 4390.723000000004, "cpu_memory_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": 3205.122000000003, "self_device_time_total_us": 3205.122000000003, "cpu_memory_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": 39.680000000000064, "self_cpu_time_total_us": 7.919999999999845, "device_time_total_us": 44.16099999999824, "self_device_time_total_us": 0.0, "cpu_memory_usage_bytes": 0, "self_cpu_memory_usage_bytes": 0, "device_memory_usage_bytes": 25460736, "self_device_memory_usage_bytes": 0, "flops": 0, "input_shapes": "[[], [], [], [], []]" }, { "key": "aten::zero_", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 26.480000000000246, "self_cpu_time_total_us": 6.048000000000002, "device_time_total_us": 44.16099999999824, "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": 20.432000000000244, "self_cpu_time_total_us": 8.512000000000398, "device_time_total_us": 44.16099999999824, "self_device_time_total_us": 44.16099999999824, "cpu_memory_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": 44.16099999999824, "self_device_time_total_us": 44.16099999999824, "cpu_memory_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": 173.23199999999997, "self_cpu_time_total_us": 20.352000000000544, "device_time_total_us": 5.5679999999993015, "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": 60.223999999999705, "self_cpu_time_total_us": 35.82399999999939, "device_time_total_us": 5.5679999999993015, "self_device_time_total_us": 5.5679999999993015, "cpu_memory_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": 28.527999999999793, "self_cpu_time_total_us": 16.751999999999953, "device_time_total_us": 4.0000000000045475, "self_device_time_total_us": 4.0000000000045475, "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": 4.0000000000045475, "self_device_time_total_us": 4.0000000000045475, "cpu_memory_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": 3.87199999999757, "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": 29.488000000000284, "self_cpu_time_total_us": 2.4160000000001673, "device_time_total_us": 3.0720000000001164, "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": 3.0720000000001164, "self_device_time_total_us": 3.0720000000001164, "cpu_memory_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": 154.1919999999999, "self_cpu_time_total_us": 8.032000000000039, "device_time_total_us": 2.495999999999185, "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": 2.495999999999185, "self_device_time_total_us": 2.495999999999185, "cpu_memory_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": 72879.795, "self_cpu_time_total_us": 72879.795, "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": 1075.264, "self_cpu_time_total_us": 1075.264, "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": 105.66399999999953, "self_cpu_time_total_us": 105.66399999999953, "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": 92.65599999999972, "self_cpu_time_total_us": 92.65599999999972, "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": 30.7199999999998, "self_cpu_time_total_us": 30.7199999999998, "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": 4566761472, "self_device_memory_usage_bytes": 4566761472, "flops": 0, "input_shapes": "[[], [], [], [], [], []]" }, { "key": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 11.328000000000202, "self_cpu_time_total_us": 11.328000000000202, "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": 11.32799999999989, "self_cpu_time_total_us": 7.471999999999582, "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": 10.351999999999862, "self_cpu_time_total_us": 7.568000000000211, "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": 9.00799999999981, "self_cpu_time_total_us": 9.00799999999981, "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": 5.344000000000051, "self_cpu_time_total_us": 3.3920000000000528, "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": 4.415999999999713, "self_cpu_time_total_us": 4.415999999999713, "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": 3.856000000000307, "self_cpu_time_total_us": 3.856000000000307, "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": 2.7839999999996508, "self_cpu_time_total_us": 2.7839999999996508, "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": 2.399999999999636, "self_cpu_time_total_us": 1.6159999999995307, "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": 1.9519999999999982, "self_cpu_time_total_us": 1.9519999999999982, "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": 1.7279999999996107, "self_cpu_time_total_us": 1.7279999999996107, "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": 1.231999999999971, "self_cpu_time_total_us": 1.231999999999971, "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, 28672]]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.7840000000001055, "self_cpu_time_total_us": 0.7840000000001055, "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": -4566763520, "self_device_memory_usage_bytes": -4566763520, "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": 5.215994315221906e-06, "p50_s": 4.656001692637801e-06, "p90_s": 7.521617226302623e-06, "p95_s": 7.528794230893254e-06, "p99_s": 7.5345358345657585e-06, "min_s": 4.511966835707426e-06, "max_s": 7.535971235483885e-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": 9.571202099323273e-06, "p50_s": 8.464005077257752e-06, "p90_s": 1.0403204942122099e-05, "p95_s": 1.4737618039362121e-05, "p99_s": 1.8205148517154158e-05, "min_s": 7.888011168688536e-06, "max_s": 1.9072031136602163e-05 }, "bias_cast": { "count": 10, "mean_s": 4.1168008465319875e-06, "p50_s": 4.039990017190576e-06, "p90_s": 4.36801346950233e-06, "p95_s": 4.4400192564353345e-06, "p99_s": 4.497623885981738e-06, "min_s": 3.840017598122358e-06, "max_s": 4.5120250433683395e-06 }, "vortex_absmax_scale": { "count": 10, "mean_s": 0.004272058588685468, "p50_s": 0.004267550975782797, "p90_s": 0.004282539803534746, "p95_s": 0.0043181654065847395, "p99_s": 0.0043466658890247345, "min_s": 0.00424494402250275, "max_s": 0.004353791009634733 }, "scale_compat_cast": { "count": 10, "mean_s": 0.001205106993438676, "p50_s": 1.369600067846477e-05, "p90_s": 0.002398188604274761, "p95_s": 0.006418293298338531, "p99_s": 0.009634377053589561, "min_s": 1.1087977327406406e-05, "max_s": 0.010438397992402315 }, "scale_to_device": { "count": 10, "mean_s": 1.4732795534655452e-05, "p50_s": 1.103198155760765e-05, "p90_s": 2.4928024504333732e-05, "p95_s": 3.0112022068351492e-05, "p99_s": 3.425922011956573e-05, "min_s": 1.0255957022309303e-05, "max_s": 3.529601963236928e-05 }, "vortex_quantize_nvfp4": { "count": 10, "mean_s": 0.006082749110646546, "p50_s": 0.00607146299444139, "p90_s": 0.006121344619896263, "p95_s": 0.006154759810306132, "p99_s": 0.006181491962634027, "min_s": 0.006053054006770253, "max_s": 0.006188175000716001 }, "params_wrap": { "count": 10, "mean_s": 1.0276801185682416e-05, "p50_s": 6.840011337772012e-06, "p90_s": 2.2163224639371033e-05, "p95_s": 2.4265589308924968e-05, "p99_s": 2.5947481044568122e-05, "min_s": 6.495974957942963e-06, "max_s": 2.636795397847891e-05 }, "activation_quantize": { "count": 10, "mean_s": 0.01161861231084913, "p50_s": 0.010411958530312404, "p90_s": 0.012876597791910168, "p95_s": 0.016871624905616035, "p99_s": 0.020067646596580745, "min_s": 0.010355374019127339, "max_s": 0.02086665201932192 }, "linear": { "count": 10, "mean_s": 0.053046907490352166, "p50_s": 0.0532312944997102, "p90_s": 0.05354207805357874, "p95_s": 0.05356652251211926, "p99_s": 0.05358607807895169, "min_s": 0.052244342979975045, "max_s": 0.05359096697065979 }, "slice_reshape": { "count": 10, "mean_s": 7.1952003054320816e-06, "p50_s": 6.583984941244125e-06, "p90_s": 9.174406295642255e-06, "p95_s": 9.707207209430634e-06, "p99_s": 1.0133447940461337e-05, "min_s": 6.320013199001551e-06, "max_s": 1.0240008123219013e-05 } }, "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": 33947648, "output_bytes": 406533120, "separate_materialization_bytes": 1795691520, "epilogue_candidate": "segmented BF16 residual gate/add" }, "profiler_summary": { "device_kernel_launches": 16, "runtime_kernel_launches": 16, "unique_device_kernels": 8, "positive_self_device_allocated_bytes": 1423510528, "negative_self_device_allocated_bytes": -1423510528, "profiled_iterations": 2, "trace_path": null, "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": 629.8879999999997, "self_cpu_time_total_us": 107.98399999999992, "device_time_total_us": 104850.064, "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": 507.5999999999997, "self_cpu_time_total_us": 417.79199999999855, "device_time_total_us": 104850.064, "self_device_time_total_us": 104846.256, "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": 104846.256, "self_device_time_total_us": 104846.256, "cpu_memory_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": 11854.592, "self_device_time_total_us": 11854.592, "cpu_memory_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": 8564.225999999995, "self_device_time_total_us": 8564.225999999995, "cpu_memory_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": 55.69599999999991, "self_cpu_time_total_us": 10.943999999999505, "device_time_total_us": 232.73600000000624, "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": 39.552000000000135, "self_cpu_time_total_us": 8.368000000000166, "device_time_total_us": 232.73600000000624, "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": 31.18399999999997, "self_cpu_time_total_us": 9.375999999999749, "device_time_total_us": 232.73600000000624, "self_device_time_total_us": 232.73600000000624, "cpu_memory_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": 232.73600000000624, "self_device_time_total_us": 232.73600000000624, "cpu_memory_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": 201.6319999999996, "self_cpu_time_total_us": 33.75999999999931, "device_time_total_us": 9.823999999983243, "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": 81.3760000000002, "self_cpu_time_total_us": 45.71200000000067, "device_time_total_us": 9.823999999983243, "self_device_time_total_us": 9.823999999983243, "cpu_memory_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": 172.0959999999998, "self_cpu_time_total_us": 10.303999999999888, "device_time_total_us": 6.783999999990556, "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": 6.783999999990556, "self_device_time_total_us": 6.783999999990556, "cpu_memory_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": 3.9040000000049986, "self_device_time_total_us": 3.9040000000049986, "cpu_memory_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": 35.136000000000195, "self_cpu_time_total_us": 18.80000000000041, "device_time_total_us": 3.808000000006359, "self_device_time_total_us": 3.808000000006359, "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": 3.808000000006359, "self_device_time_total_us": 3.808000000006359, "cpu_memory_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": 42.57600000000002, "self_cpu_time_total_us": 2.736000000000331, "device_time_total_us": 3.0399999999926877, "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": 3.0399999999926877, "self_device_time_total_us": 3.0399999999926877, "cpu_memory_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": 124293.91399999999, "self_cpu_time_total_us": 124293.91399999999, "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": 924.192, "self_cpu_time_total_us": 924.192, "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": 139.695999999999, "self_cpu_time_total_us": 139.695999999999, "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": 86.4960000000001, "self_cpu_time_total_us": 86.4960000000001, "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": 30.992000000000644, "self_cpu_time_total_us": 30.992000000000644, "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": "cuLaunchKernelEx", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 17.29599999999982, "self_cpu_time_total_us": 17.29599999999982, "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": 14.304000000000087, "self_cpu_time_total_us": 11.215999999999894, "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": 12.75200000000018, "self_cpu_time_total_us": 12.75200000000018, "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": 12.255999999999688, "self_cpu_time_total_us": 7.631999999999607, "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::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 6.400000000000318, "self_cpu_time_total_us": 3.840000000000373, "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": 5.80800000000022, "self_cpu_time_total_us": 5.80800000000022, "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": 4.6240000000000805, "self_cpu_time_total_us": 4.6240000000000805, "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::reshape", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 3.119999999999891, "self_cpu_time_total_us": 2.1920000000000073, "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": 3.088000000000193, "self_cpu_time_total_us": 3.088000000000193, "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": "[[37824, 5376], [], [], []]" }, { "key": "aten::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 2.5599999999999454, "self_cpu_time_total_us": 2.5599999999999454, "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": 2.288000000000011, "self_cpu_time_total_us": 2.288000000000011, "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": 1.3920000000000528, "self_cpu_time_total_us": 1.3920000000000528, "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::view", "count": 2, "device_type": "DeviceType.CPU", "cpu_time_total_us": 0.9279999999998836, "self_cpu_time_total_us": 0.9279999999998836, "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": 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": "[]" } ] } ] }