modelscope / modelscope/ms-swift
megatron sft下context_parallel_size无效
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 15.7k
- Forks
- 1.7k
- Avg merge
- 1d 16h
- Merged PRs (30d)
- 136
Description
做了个对照，其余参数相同，仅context_parallel_size有区别，但是显存占用无变化，疑似context_parallel_size失效。
模型为qwen3_vl_4B
case1:
在32卡L20上
--tensor_model_parallel_size 4
--pipeline_model_parallel_size 4
--context_parallel_size 2
显存占用37G
case2
在16卡L20上
--tensor_model_parallel_size 4
--pipeline_model_parallel_size 4
--context_parallel_size 1
显存占用37G
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reproducing the two reported Megatron SFT configurations for Qwen3-VL-4B, then trace where context_parallel_size is parsed and passed into the training entry point. Compare memory use and parallel-group behavior between context parallelism 1 and 2. Done means the setting has the expected effect, or a confirmed limitation is documented with a regression test.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 28/100