deepspeedai / deepspeedai/DeepSpeed

How to inference with data parallelism and model parallelism[BUG]

Open
#4,794 0 comments 0 reactions 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
I try to use inference with model parallelism and data parallelism, but it seems something wrong

logger.info("DeepSpeed Inference Initialization")
model = LlamaForCausalLM.from_pretrained(ckpt_path, device_map='cpu', torch_dtype=model_dtype)
if is_pipeline:
       model = pipeline("text-generation",
       model=model,
       tokenizer=tokenizer,
       torch_dtype=model_dtype)
       model.model = deepspeed.init_inference(
              model.model,
              mp_size = model_parallel_size,
              dtype=model_dtype,
              replace_with_kernel_inject=False
       )
       model.device = torch.device(f"cuda:{int(os.environ.get('LOCAL_RANK', 0))}")

command line: deepspeed --include localhost:4,5,6,7 --master_addr=127.0.0.1 --master_port=29501 run_inference_model_parallelism.py

The text above is main code. model_parallel_size is 2, and num_gpus is 4.

Got the warning below:
/anaconda3/lib/python3.9/site-packages/torch/distributed/distributed_c10d.py:278: UserWarning: Running all_reduce on global rank 3 which does not belong to the given group.
/anaconda3/lib/python3.9/site-packages/torch/distributed/distributed_c10d.py:278: UserWarning: Running all_reduce on global rank 2 which does not belong to the given group.

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

Start by reproducing the shown inference setup with model_parallel_size=2 on GPUs 4–7 using the provided DeepSpeed command. Inspect how ranks are assigned to the model-parallel group around the reported all_reduce warnings; done means the configuration runs without warnings and performs inference as intended.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
distributed-systems, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.