deepspeedai / deepspeedai/DeepSpeed
Multi node multi GPU sharding for inference / training Llama 405B
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Hello
We are trying to use Deepspeed to load LLama 405b across 2 nodes, of 8 x H100 SXM each.
We want to shard the model across all 16 gpus so that the model will be loaded in shards of 50.6GB each and then run inference and training with the model.
Please provide a guide on exactly how to do this, as we have been unable to figure it out.
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
No files, tests, or entry points are named. Start by investigating DeepSpeed's support for the stated two-node, 16-GPU Llama 405B inference and training setup. Done means providing a tested, exact guide for loading and sharding the model across the GPUs.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, documentation, machine-learning
- Issue type
- Documentation
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 20/100