Lightning-AI / Lightning-AI/lightning-thunder

Quantize and dequantizing model weights in trace

Open
#923 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

enhancement
Dominant language
Python
Stars
1.5k
Forks
121
PR merge metrics
No merged PRs in 30d

Description

🚀 Feature

Quantize an arbitrary model weight and have that dequantized on the fly in the traces.

Pitch

Take a list of weight names and have a corresponding transform_module and transform_initial_graph. Maybe have options for the quantization (bitsandbytes with both 8 bit and 4 bit). May also need epliogue for updating things.

Additional context

Discussed with @t-vi .

Contributor guide

No contributing guide indexed for this repository

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 transform_module and transform_initial_graph as possible entry points, but provides no files or tests. Start by locating those entry points and clarifying the weight-name inputs and quantization options; done means model weights can be quantized and dequantized on the fly in traces with the requested options covered.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
compilers, machine-learning
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.