🚨 Delete generation params from model config#41695
🚨 Delete generation params from model config#41695zucchini-nlp merged 13 commits intohuggingface:mainfrom
Conversation
|
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. |
| url = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-Audio/glass-breaking-151256.mp3" | ||
| audio, sr = librosa.load(BytesIO(urlopen(url).read()), sr=processor.feature_extractor.sampling_rate) | ||
| inputs = processor(text=prompt, audios=audio, return_tensors="pt").to(model.device) | ||
| inputs = processor(text=prompt, audio=audio, return_tensors="pt").to(model.device) |
There was a problem hiding this comment.
Audio renaming is not related, accidentally slipped from another branch. I'll rebase main after #41681 is merged
zucchini-nlp
changed the title
|
cc @vasqu maybe since Joao left and I need someone to help merge generation PRs. This one particularly is blocking me. I'd love to clean-up and break BC for v5, since generation config is too complicated at this point This PR particularly splits generation config away from model config, a dependency we had for a few ages due to legacy reasons. It's time to move on I think and raise a NOTE: bad formatted configs from hub can still be loaded for BC |
vasqu
left a comment
vasqu
left a comment
There was a problem hiding this comment.
I'm in general aligned with this, just have a few nits/questions just to be sure.
Don't forget
- the bad audio(s)
- v5 tag and add it to the issue
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
vasqu
left a comment
vasqu
left a comment
There was a problem hiding this comment.
Thx, lgtm
Time to merge with main and check conflicts 👀 just the same things as I said before re audios + v5
|
[For maintainers] Suggested jobs to run (before merge) run-slow: bart, encoder_decoder, mvp, rag, speech_encoder_decoder, udop, vision_encoder_decoder |
|
Rebased main, now the diff has no |
* i am so confused, too many circular dependencies. Delete and see what happens * pop if exists * fix a few tests * fix loading generation params from model config * oh no, revert this * replace audios with audio in docs * fix tests * fix last test * i am dumb, typo * Update src/transformers/generation/configuration_utils.py Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com> * Update tests/utils/test_modeling_utils.py Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com> --------- Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
3 participants
What does this PR do?
Disentangles config and generation config. Now the model's config will throw an error if generation params are found. To keep BC, we'll still be loading models from the hub where generation params are in config, but we'll manually drop these params from being set as an attribute.