ROCm / ROCm/aiter

Increase the Precision of fused_rope kernel

Open
#620 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
565
Forks
585
Avg merge
3d 4h
Merged PRs (30d)
366

Description

Currently, the precision for native fused rope kernel in aiter is 1e-2.
https://github.com/ROCm/aiter/blob/main/op_tests/test_rope.py.

We are working on a PR to integrate native fused rope kernel into apex: https://github.com/ROCm/apex/pull/222. The ideal precision that is expected is 1e-3 or lower.
Could a solution be found so that the precision is better?

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

Start with op_tests/test_rope.py, which documents the current 1e-2 precision for the native fused rope kernel. Compare its behavior with the 1e-3-or-lower expectation associated with the Apex integration, then verify the improved precision in that test before considering the work complete.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.