Fix shape mismatch in KyutaiSpeechToText generate() last window#46952
Merged
Conversation
Signed-off-by: jiqing-feng <[email protected]>
Signed-off-by: jiqing-feng <[email protected]>
jiqing-feng
marked this pull request as ready for review
June 29, 2026 08:44
Contributor
|
[For maintainers] Suggested jobs to run (before merge) run-slow: kyutai_speech_to_text |
Contributor
CI recapDashboard: View test results in Grafana |
Rocketknight1
approved these changes
Jul 21, 2026
Rocketknight1
left a comment
Rocketknight1
left a comment
Member
There was a problem hiding this comment.
Yep, makes sense to me, and should be safe!
Rocketknight1
enabled auto-merge
July 21, 2026 13:21
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
SangbumChoi
added a commit
to SangbumChoi/transformers
that referenced
this pull request
Jul 22, 2026
* upstream/main: (39 commits) Remove deprecated training args and `is_fast` property (huggingface#46917) Consistent output shape from `get_image_features` (huggingface#46405) Fix multi-device mxfp4 dequantization race in `_convert_moe_packed_tensors` (huggingface#47423) fix failed test cases for qwen3_omni_moe model (huggingface#47449) Fix Hunyuan-VL PIL image resize parity with reference preprocessing (huggingface#47233) Move `value` padding into the attention interfaces that need it (huggingface#47451) Simplify function dispatch for linear attention (huggingface#47450) [cache] Allow sliding window layers to be roll-backed for speculative decoding (huggingface#47447) Fix double-shifted training loss in GitForCausalLM (huggingface#47395) Fix CohereASR training-loss double-shift (same as Moonshine fix huggingface#46784) (huggingface#46895) Warn when `group_by_length` is silently ignored for iterable datasets (huggingface#47379) Update bug report list (huggingface#46607) Fix shape mismatch in KyutaiSpeechToText `generate()` last window (huggingface#46952) Optimize flash attention max seqlen computation in vision attention (huggingface#47170) fix: remove unreachable return in special token builder (huggingface#47420) Add Harry to slow CI (huggingface#47454) BLT: vectorize patch length processing (huggingface#47385) Fix `TrackioCallback` fails to log evaluation metrics after training ends (huggingface#46935) [Kimi] add integration tests (huggingface#47383) Fix typo in `MusicgenForCausalLM.generate()` (huggingface#46974) ...
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
3 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
Generating with a
KyutaiSpeechToTextForConditionalGenerationmodel crashes when thestreaming
audio_window_sizeis larger than the number of encoded audio frames:Root cause
In
prepare_inputs_for_generation, theaudio_tokensbuffer is pre-allocated withaudio_window_sizeframes. Each window the Mimi codec re-encodes the correspondingaudio slice and writes it back via
audio_tokens.copy_(new_audio_tokens). When thelast (or only) window has fewer frames than
audio_window_size, the codec emits ashorter tensor than the buffer, so the exact-size
copy_fails.Fix
Copy only the overlapping frames so the buffer and codec output stay aligned:
Single-shot generation is unchanged; repeated generation no longer crashes.
With the default
audio_window_size(estimated from the audio length) the samemismatch shows up as an off-by-one when the estimate exceeds the codec frame count
by one; the repro sets a large window to make the mismatch deterministic.
Reproduction
Before the fix it aborts with the shape mismatch above; after the fix it prints the
token count.