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

Record H3 sampler replay gate

Browse source
This commit is contained in:
Daniel Maddern 2026-08-13 15:34:13 +07:00
parent c207091825
commit 830f67ead2
1 changed files with 4 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

11
PARITY.md
View file

@ -186,7 +186,7 @@ gate and is exact.
| 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 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 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 | | 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 | | Beta/RES sampler | Yes | H3 replay is bit-exact for all 12 captured sampler calls and RES update is bit-exact from captured denoised; direct raw-output to model-wrapper denoised conversion retains small BF16/state boundary deltas |
| Video VAE decoder | Yes | No direct-versus-Comfy same-latent pixel comparison yet | | Video VAE decoder | Yes | No direct-versus-Comfy same-latent pixel comparison yet |
| Audio VAE/decode/mux | No | Preview intentionally produces video only | | Audio VAE/decode/mux | No | Preview intentionally produces video only |
| Full Ref2VA | No | References, vision conditioning, VAE encode, audio, and muxing are unimplemented | | Full Ref2VA | No | References, vision conditioning, VAE encode, audio, and muxing are unimplemented |
@ -198,15 +198,12 @@ unless the checkpoint, Comfy version, prompt, or backend changes.
1. **Final video latent layout.** Compare direct `unpatchify_video` output 1. **Final video latent layout.** Compare direct `unpatchify_video` output
with Comfy before entering the sampler. with Comfy before entering the sampler.
2. **Sampler replay using Comfy state.** Feed `initial.pt` and captured 2. **VAE pixel gate.** Decode the identical captured final video latent in
`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.
3. **VAE pixel gate.** Decode the identical captured final video latent in
direct and Comfy and compare pixels before ffmpeg. This decides whether any direct and Comfy and compare pixels before ffmpeg. This decides whether any
residual grid comes from decoder behavior. residual grid comes from decoder behavior.
4. **End-to-end FL2VA preview.** Generate the reference dragon only after 3. **End-to-end FL2VA preview.** Generate the reference dragon only after
gates 1-8 pass. Compare its raw final latent first, then video. gates 1-8 pass. Compare its raw final latent first, then video.
5. **Feature/performance work.** Only then add audio, Ref2VA/reference paths, 4. **Feature/performance work.** Only then add audio, Ref2VA/reference paths,
Sage3, CUDA graphs, and multi-GPU execution. Sage3, CUDA graphs, and multi-GPU execution.
## Existing Tools And Their Intended Gate ## Existing Tools And Their Intended Gate