max_new_tokens as a parameter to answer_question() and default=256 instead of 512 #86
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
max_new_tokens is added as a parameter to answer_question() and default is set to 256 instead of 512. Reasons: 1) more control, 2) at least in test generations I found about 200 wasn't interrupted (100 was small) ; 3) in order to reduce the delay when the model enters in loops generating .............. and taking a lot of time. Also: can some sort of an interrupt be added, either with a callback like in GPT4All, also a constraint when capturing a repetition of tokens etc., repetition_penalty etc.? Can we add a streaming mode? Thanks.
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Outdated suggestions cannot be applied.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.
max_new_tokens is added as a parameter to answer_question() and default is set to 256 instead of 512.
Reasons:
Also, for avoiding the repetition loops, can some sort of an interrupt be added, as a callback like in GPT4All and a constraint when capturing repetition of tokens, unusual sequences like these dots etc.: repetition_penalty etc.? Also: a streaming mode? (I had a glance on the code, I may try to figure it out.)
Sample repetition loop:
Z:\LMSYS-Free-GPT4-Claude-15-4-2024-2024-04-15_18-33-46.mp4_snapshot_05.28.266.png
The image is a screenshot of a website called "LMSys Chatbot Arena". The website is predominantly white with blue and orange accents. The main focus is a banner at the top of the screen, which is white with red and blue text. The text reads "Free GPT4, claude, llama, code llama, mixral-of-experts, command-r-plus..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
The picture itself had "..." in the text.