ROCm / ROCm/AMDMIGraphX

Efficiently support mixed precision w8a16 models

Open
#3,836 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
333
Forks
150
Avg merge
4d 19h
Merged PRs (30d)
54

Description

Generalize the work to efficiently support w4a16, to support w8a16 (weights in an 8 bit format, e.g. int8, eventually fp8); activations in fp16 (eventually also bf16).

"Efficiently" here means that the weights will stay as int8, and they will be dequantized only in the context of the kernel that will use them; just like how it works for int4 weights.

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

No files, tests, or entry points are named. Start by locating the existing int4 and w4a16 handling, then trace how weights are represented and dequantized within the consuming kernel. Done means the approach is generalized to efficiently support w8a16 weights with fp16 activations while preserving the stated in-kernel dequantization behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
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.