Add missing ckpt in config docs#16900
Conversation
|
The documentation is not available anymore as the PR was closed or merged. |
| instantiate an Speech2Text2 model according to the specified arguments, defining the model architecture. | ||
| Instantiating a configuration with the defaults will yield a similar configuration to that of the Speech2Text2 | ||
| [facebook/s2t-small-librispeech-asr](https://huggingface.co/facebook/s2t-small-librispeech-asr) architecture. | ||
| [facebook/s2t-wav2vec2-large-en-de](https://huggingface.co/facebook/s2t-wav2vec2-large-en-de) architecture. |
There was a problem hiding this comment.
facebook/s2t-small-librispeech-asr has model_type": "speech_to_text, but here is for the model Speech2Text2 (i.e. speech_to_text_2)
sgugger
left a comment
There was a problem hiding this comment.
Thanks for fixing all of those! It would be awesome to have some kind of quality script to check we don't introduce new faulty checkpoints.
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Yes, I do have some (draft) check locally. I plan to add it in another PR (unless it's necessary to do so in this PR). |
src/transformers/models/squeezebert/configuration_squeezebert.py
Outdated
Show resolved
Hide resolved
src/transformers/models/xlm_roberta/configuration_xlm_roberta.py
Outdated
Show resolved
Hide resolved
|
Thank you @NielsRogge I should try to use the correct names, as defined in |
There was a problem hiding this comment.
Thanks a lot for this PR, awesome that this gets improved.
Left some comments, just for consistency, I would always use the template:
"will yield a similar configuration of that of the - snake-cased model name - checkpoint name architecture".
I will add this to the check I currently have (locally, but will push to another PR), thanks! |
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
patrickvonplaten
left a comment
There was a problem hiding this comment.
Looked at all the speech models - looks good to me!
|
Merge now. Thanks for the review. With this PR, all configs are good except the following (which are expected, since those composite models don't have full default config arguments - they rely on the encoder and decoder configs.)
|
* add missing ckpt in config docs * add more missing ckpt in config docs * fix wrong ckpts * fix realm ckpt * fix s2t2 * fix xlm_roberta ckpt * Fix for deberta v2 * Apply suggestions from code review Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * use only one checkpoint for DPR * Apply suggestions from code review Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> Co-authored-by: ydshieh <ydshieh@users.noreply.github.com> Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
What does this PR do?
As discussed on Slack, I worked on the
Configfiles to add missing information about checkpoints, or correct them.Instantiating a configuration with the defaults will yield a similar configuration to that of the Speech2Text2 [mentioned checkpoint]hidden_dim,num_layersmight be differentsimilar, so I think it is fine (..?)@patrickvonplaten Could you take a look on the speech models?
@NielsRogge Could you take a look on the vision models?