ROCm / ROCm/AMDMIGraphX

Fast tuning mode

Open
#1,396 0 comments 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

Presently, miopen tuning can take hours. While this produces high quality results, ideally we should have a mode where only dynamic kernels are considered (i.e. without any kernel compilation in miopen; aka "hybrid find").

If such a feature is available, then migraphx could inform clients (such as the UIF pruner) about performance estimate of the model, with the proposed pruned dimensions.

A further possible improvement could just consider tuning parameters of the modified (pruned) problem in the neighborhood of the tuning parameters of the unmodified (unpruned) problem, for the same solver.

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 by locating the MIOpen tuning integration and the path that selects kernels for model inference. Clarify whether the goal is a dynamic-kernel-only hybrid-find mode, neighborhood tuning, or both, and define how performance estimates would be exposed to the UIF pruner. Done should include an agreed scope and validation of the estimate 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.