diff --git a/tools/compare_fl2va_steps.py b/tools/compare_fl2va_steps.py index f70ff60..08ea018 100644 --- a/tools/compare_fl2va_steps.py +++ b/tools/compare_fl2va_steps.py @@ -7,9 +7,7 @@ import torch from h3_blackwell_runtime.checkpoint import H3Checkpoint from h3_blackwell_runtime.denoiser import H3PackedDenoiser from h3_blackwell_runtime.packing import H3PromptPacker, unpatchify_video -from h3_blackwell_runtime.qwen3vl_text import Qwen3VLPromptConditioner from h3_blackwell_runtime.sampler import _audio_sigma, _unpack_audio, res_multistep_update -from h3_blackwell_runtime.token_refiner import H3TokenRefiner root = "/artifacts/fl2va-sampler-reference" @@ -17,7 +15,8 @@ initial = torch.load(f"{root}/initial.pt", map_location="cuda", weights_only=Fal steps = [torch.load(path, map_location="cuda", weights_only=False) for path in sorted(glob.glob(f"{root}/step_*.pt"))] sigmas = initial["sigmas"].to("cuda") checkpoint = H3Checkpoint("/models/minimax_h3_fl2va_pruned_nvfp4.safetensors") -text = H3TokenRefiner(checkpoint)(Qwen3VLPromptConditioner("/text-encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors")("A brass-and-paper dragon flies above a rain-washed old city at blue hour.")) +captured_input = torch.load("/artifacts/capture/input.pt", map_location="cuda", weights_only=False) +text = captured_input["hidden"][:17].unsqueeze(0) model = H3PackedDenoiser.from_checkpoint(checkpoint, attention_backend="sage2").eval() packer = H3PromptPacker(checkpoint)