lm-sys / lm-sys/FastChat

Why `model_max_length` from vicuna 1.1 and vicuna 1.3 are different?

Open
#1,876 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
39.5k
Forks
4.8k
PR merge metrics
No merged PRs in 30d

Description

I tested vicuna 1.1 on text-generation-web-ui and the model can handle long input even bigger than `max_position_embeddings`. Now, in vicuna 1.3, I kept getting warning when my input is longer than 2048 tokens. When I check the `tokenizer_config.json`, vicuna 1.1 has a very large value for `model_max_length` which is `1000000000000000019884624838656` rather than vicuna 1.3 which the value is `2048`.

Why is this happening? Is this intentional? What could be the problem if I somehow change the value manually?

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by comparing the Vicuna 1.1 and 1.3 tokenizer_config.json values with their max_position_embeddings settings, then trace how text-generation-web-ui uses model_max_length for input warnings. Done means documenting whether the difference is intentional and the consequences of manually changing the value.

Written by the indexing model from the issue text.

Assessment

Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.