Conversation pipeline fixes #26795
Merged
Conversation pipeline fixes #26795
Conversation
This file contains 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
|
The documentation is not available anymore as the PR was closed or merged. |
|
Also cc @lewtun - now you should actually be able to just use this pipeline in the docstrings instead of needing to do it manually in the docstrings with |
ArthurZucker
approved these changes
Oct 16, 2023
| n = model_inputs["input_ids"].shape[1] | ||
| if max_length - minimum_tokens < n: |
| @@ -268,6 +268,10 @@ def __call__(self, conversations: Union[Conversation, List[Conversation]], num_w | |||
| # Otherwise the threads will require a Conversation copy. | |||
| # This will definitely hinder performance on GPU, but has to be opted | |||
| # in because of this BC change. | |||
| if isinstance(conversations, list) and isinstance(conversations[0], dict): | |||
There was a problem hiding this comment.
can we update the doc to mention the type of input we expect?
gante
approved these changes
Oct 16, 2023
| conversation = model_inputs.pop("conversation") | ||
| generate_kwargs["max_length"] = max_length | ||
| if "max_length" not in generate_kwargs and "max_new_tokens" not in generate_kwargs: | ||
| generate_kwargs["max_new_tokens"] = 256 |
There was a problem hiding this comment.
👍
It should be safe to add this default, as generate should throw a warning if the generation goes beyond what's supported by the model.
Rocketknight1
force-pushed
the
conversation_pipeline_fixes
branch
from
d9d9dea to
89e48f6
Compare
EduardoPach
pushed a commit
to EduardoPach/transformers
that referenced
this pull request
Nov 19, 2023
* Adjust length limits and allow naked conversation list inputs * Adjust length limits and allow naked conversation list inputs * Maybe use a slightly more reasonable limit than 1024 * Skip tests for old models that never supported this anyway * Cleanup input docstrings * More docstring cleanup + skip failing TF test * Make fixup
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This PR makes a couple of fixes to
ConversationalPipelineto make it a lot easier to use:Conversationclass is quite hard for users to discover, and this is a lot more intuitive.max_lengthbecause very few models set this parameter, and so it's almost always the defaultPretrainedConfigvalue of 20, which is very low. Before this change, most calls toConversationalPipelineproduced no output or unnecessarily truncated the input because this limit was hit. We change the pipeline to usemax_new_tokensinstead, which is more modern.cc @ArthurZucker for pipeline review and @gante if he has any comments about setting the generation parameters properly!