NVIDIA / NVIDIA/nvbench

Smart selection of the entropy step

Open
#154 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Cuda
Stars
927
Forks
123
Avg merge
2d 10h
Merged PRs (30d)
2

Description

The custom stopping criterion introduced in https://github.com/NVIDIA/nvbench/pull/151 currently triggers linear regression and R2 computation every second is_finished() call. Per @julie-is-late comment:

depending on the expected amount of time a test can take, i.e. if a test is very fast when compared to the amount of time to compute entropy, it might be better to calculate entropy and check if you're done at even bigger steps. This could even then be made "smart" by taking into account: how far is the current slope from the goal angle? what is the cuda_time of each run?

We should experiment with this idea.

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 the custom stopping criterion introduced in pull request #151 and trace its is_finished() calls, linear regression, and R2 computation. Compare the current fixed cadence with the proposed use of slope distance from the goal angle and each run's cuda_time; done means an experimentally supported smart entropy-checking strategy is defined.

Written by the indexing model from the issue text.

Assessment

Domain
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.