daniel/h3-blackwell-runtime
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Capture H3 denoiser calls per sampler step

Browse source
This commit is contained in:
Daniel Maddern 2026-08-13 15:23:18 +07:00
parent 889e76b85b
commit e679178bf8
1 changed files with 10 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
tools/patch_comfy_h3_capture.py
View file

@ -14,24 +14,29 @@ def replace_once(path, old, new):
model = Path("/opt/ComfyUI/comfy/ldm/minimax/model.py")
replace_once(model, "import math\n", "import math\nimport os\n")
replace_once(model, "VISUAL_COND_TIMESTEP = 0.999\n", "VISUAL_COND_TIMESTEP = 0.999\nH3_CAPTURE_ACTIVE = False\n")
replace_once(model, "VISUAL_COND_TIMESTEP = 0.999\n", "VISUAL_COND_TIMESTEP = 0.999\nH3_CAPTURE_ACTIVE = False\nH3_CAPTURE_INDEX = 0\n")
replace_once(
model,
" # blocks\n patches_replace = transformer_options.get(\"patches_replace\", {})\n",
" # Capture a single fully assembled payload before the first transformer block.\n"
" global H3_CAPTURE_ACTIVE\n"
" # Capture fully assembled payloads before each transformer call.\n"
" global H3_CAPTURE_ACTIVE, H3_CAPTURE_INDEX\n"
" capture_dir = os.getenv(\"H3_CAPTURE_DIR\")\n"
" if capture_dir and not H3_CAPTURE_ACTIVE:\n"
" H3_CAPTURE_ACTIVE = True\n"
" os.makedirs(capture_dir, exist_ok=True)\n"
" torch.save({\"hidden\": h.detach().cpu(), \"timesteps\": t_vals.detach().cpu(), \"t_emb\": t_emb.detach().cpu(), \"position_ids\": layout.position_ids, \"segments\": mod_segments, \"video_x\": video_x.detach().cpu(), \"audio_x\": audio_x.detach().cpu()}, os.path.join(capture_dir, \"input.pt\"))\n\n"
" payload = {\"hidden\": h.detach().cpu(), \"timesteps\": t_vals.detach().cpu(), \"t_emb\": t_emb.detach().cpu(), \"position_ids\": layout.position_ids, \"segments\": mod_segments, \"video_x\": video_x.detach().cpu(), \"audio_x\": audio_x.detach().cpu()}\n"
" torch.save(payload, os.path.join(capture_dir, \"input.pt\"))\n"
" torch.save(payload, os.path.join(capture_dir, f\"input_{H3_CAPTURE_INDEX:02d}.pt\"))\n\n"
" # blocks\n patches_replace = transformer_options.get(\"patches_replace\", {})\n",
)
replace_once(
model,
" return [-video_out.to(video_x.dtype), -audio_out.to(audio_x.dtype)]\n",
" if capture_dir and H3_CAPTURE_ACTIVE:\n"
" torch.save({\"video\": video_out.detach().cpu(), \"audio\": audio_out.detach().cpu(), \"video_segment\": video_seg, \"audio_segment\": audio_seg, \"adaln_t_table\": self.adaln_t_table.detach().cpu()}, os.path.join(capture_dir, \"output.pt\"))\n"
" payload = {\"video\": video_out.detach().cpu(), \"audio\": audio_out.detach().cpu(), \"video_segment\": video_seg, \"audio_segment\": audio_seg, \"adaln_t_table\": self.adaln_t_table.detach().cpu()}\n"
" torch.save(payload, os.path.join(capture_dir, \"output.pt\"))\n"
" torch.save(payload, os.path.join(capture_dir, f\"output_{H3_CAPTURE_INDEX:02d}.pt\"))\n"
" H3_CAPTURE_INDEX += 1\n"
" H3_CAPTURE_ACTIVE = False\n"
" return [-video_out.to(video_x.dtype), -audio_out.to(audio_x.dtype)]\n",
)