[bugfix]: read H3 geometry from checkpoint JSON so lazy load can drop the DiT before VAE decode - #2
Merged
Satyam-53 merged 2 commits intoAug 31, 2026
Conversation
… the DiT before VAE decode Input prep and unpatchify were holding live VAE/DiT proxies just for two integers, which loaded the video VAE before Qwen and kept the DiT resident through decode. Auto-enable --lazy-module-load on unified memory and on single-GPU FastH3 examples.
…ing docs The MLX phase-order sentence and "start a new generator" instruction belong to a different design. A later generate() reloads in-process.
Satyam-53
approved these changes
Aug 31, 2026
|
/merge |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
spatial_compression_ratio,latent_channels, andpatch_size. Those scalars now come from checkpointconfig.jsonviaupdate_model_arch(no weights).--lazy-module-loadconfirmed that order (VAE 10.34 GiB after DiT release of 65.51 GiB).--lazy-module-loadis auto-on for unified memory (Nonedefault) and forbasic_fasth3.py/basic_minimax_h3_t2v.pywhen--num-gpus 1.--no-lazy-module-loadstill keeps every component resident. Docs now describe the opt-in flag instead of always-on sequential start.This is the review follow-up for hao-ai-lab/FastVideo#1761. It does not include CUDA TAEH3.
Test plan
pytestlazy-load, H3 VAE streaming, unified-memory policy, FastH3 example, API parser (94 passed)--lazy-module-load, full VAE: geometry from config with no VAE weights at input prep; Qwen → release → DiT denoise ~138 s → DiT release → VAE 10.34 GiB decodelazy-module-load/ [perf]: MiniMax H3 on GB10 - load pipeline components on demand instead of all at once hao-ai-lab/FastVideo#1761