deepspeedai / deepspeedai/DeepSpeed
[BUG] Jamba (Mamba+MoE) + ZeRO3 + LoRA training hangs
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Model Link: https://huggingface.co/ai21labs/Jamba-v0.1
Reproduce script in OpenRLHF: https://github.com/OpenLLMAI/OpenRLHF/blob/main/examples/scripts/train_sft_jamba_lora.sh
Please pip install mamba-ssm causal-conv1d>=1.2.0
and set --micro_train_batch_size 1 to reproduce this bug ( IMPORTANT! )
--micro_train_batch_size 4 works well
Mixtral + ZeRO3 + LoRA works well with the same hyperparameters (`--micro_train_batch_size 1):
see https://github.com/OpenLLMAI/OpenRLHF/blob/main/examples/scripts/train_sft_mixtral_lora.sh
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 with OpenRLHF's examples/scripts/train_sft_jamba_lora.sh and reproduce the hang using Jamba with ZeRO3, LoRA, and --micro_train_batch_size 1 after installing mamba-ssm and causal-conv1d>=1.2.0. Compare the working Mixtral script and the batch-size-4 behavior; done means Jamba training no longer hangs with batch size 1.
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
- 35/100