lm-sys / lm-sys/FastChat

Loading vicuna-13b-1.5 with single GPU in 16GB VM triggers system OOM

Open
#2,359 1 comment 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

It seems that due to the way load_model() handles cuda devices, when running a worker with --num-gpus 1nodevice_map`` is set, which causes my VM to OOM:

[23502.263474] python3 invoked oom-killer: gfp_mask=0x1100cca(GFP_HIGHUSER_MOVABLE), order=0, oom_score_adj=0
[23502.263485] CPU: 5 PID: 1215 Comm: python3 Tainted: P           OE     5.15.0-82-generic #91-Ubuntu

while loading shards:

Loading checkpoint shards:  33%|█████████████████████████████████                                                                  | 1/3 [00:14<00:28, 14.31s/it]
Killed

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 tracing the load_model() path used when a worker is started with --num-gpus 1, focusing on how CUDA devices and device_map are selected. Reproduce the single-GPU load with vicuna-13b-1.5 in a 16GB VM and verify that checkpoint loading completes without a system OOM.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
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.