deepspeedai / deepspeedai/DeepSpeed

AutoModelFromCausalLLM of Bloom not releasing GPU memory after each inference batch [BUG]

Open
#3,508 1 comment 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

bug inference
Dominant language
Python
Stars
43.1k
Forks
5k
Avg merge
4d 15h
Merged PRs (30d)
112

Description

Describe the bug
Hi there, I have set torch.no_grad() and torch.cuda.empty_cache(), but the GPU still encounters out-of-memory (OOM) errors after a few inferences. My torch version is 1.13.1, deepspeed version is 0.9, and transformer version is 4.28, cuda driver 11.6 with v100

Expected behavior
auto release the memory
ds_report output
all good

Contributor guide

Open the contributing guide

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

No file or test is named; start at the AutoModelFromCausalLLM Bloom inference entry point and reproduce the reported GPU memory growth across batches with the stated PyTorch, DeepSpeed, Transformers, and CUDA versions. Compare allocated memory after each batch and define done as repeated inference without progressive growth or an OOM.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning, performance
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.