deepspeedai / deepspeedai/DeepSpeed

[BUG] some question about flops_profiler

Open
#4,593 1 comment 0 reactions 1 assignee View on GitHub

@ShijieZZZZ is already working on this.

Since Nov 6, 2023.

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

Description

Describe the bug
question:

  1. when I use flops_profiler to profile TFLOPS of Megatron GPT, I get the value not correct.
  2. when I use flops_profiler to train LLAMA2, I can't get any infos.

To Reproduce
Steps to reproduce the behavior:
for question 1:

  1. go to the project https://github.com/microsoft/Megatron-DeepSpeed,
  2. modify the shell script https://github.com/microsoft/Megatron-DeepSpeed/blob/main/examples_deepspeed/run_deepspeed_example.sh, add "flops_profiler" in ds_config, then I get the different value from which Megatron returns to me.
    for question 2:
  3. go to the project https://github.com/hiyouga/LLaMA-Factory/tree/main
  4. use deepspeed for training llama v2 and add "flops_profiler" in ds_config. I can't get any info about TFLOPS

Expected behavior

  1. get the same TFLOPS as Megatron returns to me
  2. get some info when I am to train the llama v2

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.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.