deepspeedai / deepspeedai/DeepSpeed

[AutoTP + DS2] no memory reduction when using auto = 4

Open
#7,684 7 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

I trained model qwen2.5-7B in GPU with llamafactory deepspeed zero2 + autotp. And there is no obvious memory reduction.

When individually using zero2, the average memory of 8 cards is 42g, while the average of zero2+autotp=4 is also nearly 42.

Here is my deepspeed config.


{
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
  "gradient_accumulation_steps": "auto",
  "gradient_clipping": "auto",
  "zero_allow_untested_optimizer": true,
  "fp16": {
    "enabled": "auto",
    "loss_scale": 0,
    "loss_scale_window": 1000,
    "initial_scale_power": 16,
    "hysteresis": 2,
    "min_loss_scale": 1
  },
  "bf16": {
    "enabled": "auto"
  },
  "zero_optimization": {
    "stage": 2,
    "allgather_partitions": true,
    "allgather_bucket_size": 5e8,
    "overlap_comm": false,
    "reduce_scatter": true,
    "reduce_bucket_size": 5e8,
    "contiguous_gradients": true,
    "round_robin_gradients": true
  },
  "tensor_parallel":{
      "autotp_size": 4
  }
}

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 reported comparison with the supplied DeepSpeed configuration: ZeRO stage 2 alone versus AutoTP with autotp_size 4, using the same Qwen2.5-7B training setup. Measure per-GPU memory in both runs and trace the AutoTP and ZeRO integration to identify why memory remains near 42 GB. Done means the cause is isolated and the combined configuration shows the expected reduction or the limitation is documented.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
distributed-systems, machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.