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

Match H3 final output heads

Browse source
This commit is contained in:
Daniel Maddern 2026-08-13 13:09:42 +07:00
parent a85fe116c1
commit 985a529fc9
2 changed files with 5 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
PARITY.md
View file

@ -184,6 +184,7 @@ gate and is exact.
| Token refiner | Yes | Bit-exact from captured 5376-wide refiner input through both blocks and final RMSNorm; Qwen-to-refiner projection boundary is still not separately captured |
| Prompt-only FL2VA packer | Yes | Bit-exact for the coherent captured text-only FL2VA DiT input; no keyframe/reference condition rows |
| H3 DiT backbone | Yes | Bit-exact through all 50 blocks from the coherent assembled FL2VA input; requires the standalone Comfy Kitchen fused Q/K RMSNorm + split-half RoPE operator |
| H3 final layer | Yes | Bit-exact final AdaLN, target-row modulation, and video/audio patch rows; Comfy materializes the AdaLN and output-head biases through BF16 |
| H3 DiT | Yes | Strict all-block numeric parity not achieved |
| Beta/RES sampler | Yes | Exact sigma/state/update parity not demonstrated; direct preview has wrong initial audio state |
| Video VAE decoder | Yes | No direct-versus-Comfy same-latent pixel comparison yet |
@ -195,10 +196,8 @@ gate and is exact.
Only these are outstanding. Do not recapture or revisit rows marked complete
unless the checkpoint, Comfy version, prompt, or backend changes.
1. **Final rows and video latent.** Final DiT hidden state, final AdaLN, and
final RMSNorm are exact. Repair the remaining final target-segment
modulation/FP32-head boundary (`video_hidden` max `0.00598395`) before
`unpatchify_video` output with Comfy before entering the sampler.
1. **Final video latent layout.** Compare direct `unpatchify_video` output
with Comfy before entering the sampler.
2. **Sampler replay using Comfy state.** Feed `initial.pt` and captured
`sigmas` directly to the sampler. Compare video and audio denoised/update
states for all 12 saved steps. Do not test seed equality until this passes.

4
src/h3_blackwell_runtime/final.py
View file

@ -45,9 +45,9 @@ class H3FinalLayer(nn.Module):
checkpoint.tensor("final_layer.adaln_proj.linear.weight", dtype=torch.bfloat16),
checkpoint.tensor("final_layer.adaln_proj.linear.bias", dtype=torch.bfloat16),
checkpoint.tensor("final_layer.video_out.weight", dtype=torch.bfloat16),
checkpoint.tensor("final_layer.video_out.bias", dtype=torch.float32),
checkpoint.tensor("final_layer.video_out.bias", dtype=torch.bfloat16),
checkpoint.tensor("final_layer.audio_out.weight", dtype=torch.bfloat16),
checkpoint.tensor("final_layer.audio_out.bias", dtype=torch.float32),
checkpoint.tensor("final_layer.audio_out.bias", dtype=torch.bfloat16),
)
def forward(