lm-sys/FastChat

[BUG] RWKV Models are not configured to use Cuda GPU lists.

Open

#1,248 opened on May 15, 2023

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

Description

Hello,

The file (fastchat/model/rwkv_model.py) has a RWKV stratergy (strategy="cuda fp16") being set that is not imlemented to use the native --gpus or --num-gpus fschat command arguments.

Example, Launching a model worker with (--num-gpus 4 --gpus 0,1,2,3) does not spawn the RWKV instance with these GPU specifications and just uses GPU 1

strategy="cuda fp16"

VS.

strategy="cuda:0 fp16 -> cuda:1 fp16 -> cuda:2 fp16 -> cuda:3 fp16 -> cuda:4 fp16"

Contributor guide

Tech stack
python
Domain
backendmachine learning
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
2
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
fresh
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
Basic PythonUnderstanding of CUDA GPU management
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40
Research direction
The issue is in fastchat/model/rwkv model.py where the strategy is hardcoded as 'cuda fp16'. The model worker command line arguments num gpus and gpus are not being used to set the strategy accordingly. The fix should parse these arguments and construct a strategy string that specifies which GPUs to use, e.g., 'cuda:0 fp16 > cuda:1 fp16'. Refer to the model worker's argument parsing in fastchat/serve/model worker.py and the existing GPU allocation logic.