Lightning-AI / Lightning-AI/litgpt

Faster inference, Paged attention from vllm

Open
#224 2 comments 2 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

documentation
Dominant language
Python
Stars
13.7k
Forks
1.5k
Avg merge
15h 37m
Merged PRs (30d)
1

Description

I'm having great qualitative results from Falcon finetuned with adaptersv2.

The inference is better than what I have with huggingface/peft and lora, but still slow for scaling up.

Could the ideas or code from Paged attention https://github.com/vllm-project/vllm be used to really speed up the inference with parallel sampling and larger batch sizes?

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

The issue names no files, tests, or entry points. Start by locating litgpt's inference and sampling paths, then compare them with vLLM's PagedAttention approach. Done would require an agreed implementation scope and benchmarks showing improved parallel sampling or larger batch-size performance.

Written by the indexing model from the issue text.

Assessment

Tech stack
huggingface, python
Domain
ai, machine-learning, performance
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.