lm-sys/FastChat

support for 4bit quantization from transfomer library.

Open

#1,798 opened on Jun 27, 2023

View on GitHub
 (7 comments) (2 reactions) (0 assignees)Python (38,959 stars) (4,736 forks)batch import
enhancementgood first issue

Description

Loading a vicuna13B using 4bit quantization from the transformers library is possible load_in_4bit. How difficult could be for Fastach to support it?

Contributor guide

Tech stack
pythonpytorch
Domain
machine learningai
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
needs maintainer response
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
PythonPyTorchHugging Face TransformersQuantization basics
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
50
Research direction
Investigate how FastChat currently loads models (e.g., in model worker.py) and replicate the load in 4bit parameter from the transformers library. Review the transformers quantization documentation linked in the issue to understand the API. Check existing comments for any insights or blockers. Implement the parameter in the model loading pipeline, ensuring compatibility with the existing model serving infrastructure.