deepspeedai / deepspeedai/DeepSpeed

LLama factory, quantized model and deepspeed compatibility

Open
#7,096 0 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

hello,
May I ask at this moment if deepspeed is compatible with 4-bit quantized model at ZeRo-3(multi-GPUs)?
I downloaded a Deepseek-32B-4bit model and try to use LLama factory to launch Lora finetuning, and was prompted with the following error:

main/src/llamafactory/model/model_utils/quantization.py", line 114, in configure_quantization
[rank1]: raise ValueError("DeepSpeed ZeRO-3 or FSDP is incompatible with PTQ-quantized models.")
[rank1]: ValueError: DeepSpeed ZeRO-3 or FSDP is incompatible with PTQ-quantized models.

If LLama factory is incompatible, can I use python scripts to launch finetuning with deepspeed Zero-3?

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 with main/src/llamafactory/model/model_utils/quantization.py at line 114 and trace the reported ZeRO-3 or FSDP compatibility check. Compare the Deepseek-32B 4-bit LoRA fine-tuning setup with the requested multi-GPU workflow; done would require a documented compatibility answer or a clearly supported launch path.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
distributed-systems, machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
15/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.