-
Notifications
You must be signed in to change notification settings - Fork 25.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
BlenderbotSmall incorrect usage of start and end tokens #22301
Comments
Hey! Thanks for reporting, will investigate! |
Hey! When I use the Conversational pipeline I get the same outputs as you. In [35]: tokenizer.encode("<s>")
Out[35]: [3, 330, 1360]
In [36]: tokenizer.encode("__start__")
Out[36]: [1] The doc example should be updated, or the tokenizer only should be updated. |
Just bumping this again (in response to being marked as stale) |
When I checked the original PR that added BlenderBot (could not really find anyting on the original repo ... ) seems like the doc example should be updated to use |
Closed in #24092 |
System Info
transformers
version: 4.27.2Who can help?
@ArthurZucker @younesbelkada @Narsil
Information
Tasks
examples
folder (such as GLUE/SQuAD, ...)Reproduction
As stated in the documentation: https://huggingface.co/docs/transformers/model_doc/blenderbot-small#transformers.BlenderbotSmallForConditionalGeneration.forward.example
the model should use
</s>
and<s>
for separating the user input and response:However, these tokens are not present in the vocabulary or special tokens
I assume they should be replaced with
__start__
and__end__
?I have also tried to use the ConversationPipeline, and follow steps outlined here, but I always get nonsensical results.
Even when trying the hosted inference API for the model (https://huggingface.co/facebook/blenderbot_small-90M), it either repeats itself, or doesn't follow in conversation.
Expected behavior
The tokens should be correct, and the chatbot should engage in more realistic conversation
The text was updated successfully, but these errors were encountered: