deepspeedai / deepspeedai/DeepSpeed
Model saved from deepspeed and accelerate cannot be loaded or incomeplete
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
I train/fine-tune the Mistral-24B small model (mistralai/Mistral-Small-24B-Instruct-2501) using deepspeed and accelerate
and I saved the model using following commands:
if accelerator.is_main_process:
model =accelerator.unwrap_model(trainer.model)
model.save_pretrained(model_path)
tokenizer.save_pretrained(model_path)
However when I try to load the model using
AutoModelForCausalLM.from_pretrained(model_path)
I got weight mismatch size error.
This is how the folder look like can
folder/
chat_template.jinja
config.json
generate_config.json
model.safetensors
special_tokens_map.json
tokenizer_config.json
tokenizer.json
Can anyone tell me what is going on here and how to solve the problem.
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 save flow using DeepSpeed and Accelerate, including accelerator.unwrap_model(trainer.model), model.save_pretrained(model_path), and AutoModelForCausalLM.from_pretrained(model_path). Compare the saved folder contents and reported weight-mismatch sizes to determine whether the model can be loaded completely; done means the saved Mistral-Small-24B-Instruct-2501 model loads successfully.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 32/100