[feat]: Add opt-in CUDA TAEH3 preview decode for FastH3 - #1
Closed
aryan5v wants to merge 1 commit into
Closed
Conversation
Skip the 9.7 GiB video VAE on T2VA so GB10 can reconstruct alpine 768 in a couple of seconds instead of a full VAE pass.
Author
|
This was opened against CUDA TAEH3 is now an independent PR against official Closing this one as superseded. |
Author
|
Superseded by hao-ai-lab#1795 (against main, not 1761). |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
--video-decode-backend taeh3on the CUDA FastH3 path. The full MiniMax video VAE stays the default.lazy-module-load). [feat] Add an optional MLX TAEH3 preview decoder hao-ai-lab/FastVideo#1794 remains the MLX TAEH3 PR and is unchanged.Architecture and weights are Ollin Boer Bohan's MIT-licensed TAEH3 (pinned SHA-256
4fd022bfcab08772fe0536b17ea1a3bbb5625be11e397868d1c5d891863d4c13). This is approximate preview reconstruction, not lossless.Spark GB10
Same alpine prompt, VSA-DataFree, 768×1344×124, seed 2026,
--lazy-module-load --video-decode-backend taeh3 --vsa-kernel triton --no-fa4 --no-warmup:1761 plus the full VAE still OOMs that canvas. TAEH3 is what lets 768 complete here.
Test plan
pytest fastvideo/tests/vaes/test_minimax_h3_taeh3.py fastvideo/tests/inference/test_basic_fasth3_profile.py fastvideo/tests/api/test_schema_parity_inventory.py::test_fastvideo_args_fields_are_classifiedfasth3_all_run_01.mp4; log showsMiniMax-H3 video decode: TAEH3 previewand noLoaded deferred module vae--video-decode-backend h3-vaeis unchanged