stan-dev / stan-dev/cmdstanr

LLM code optimizations

Open
#1,220 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

internal-code
Dominant language
R
Stars
160
Forks
69
Avg merge
1d 19h
Merged PRs (30d)
15

Description

Debrief is a recent package which is (broadly) aimed to provide LLM's profiling information in a clear manner. They have a case study describing major performance benefits they observed by setting an LLM loose on a problem. I tried something similar in one of my packages and am looking to get some nice performance benefits (including decreased memory usage not captured by touchstone).

My prompt was more minimal than theirs: "I want you to use the profvis and debrief package to iteratively make improvements to the main, scanning functions. The workflow: profile with profvis, analyze with pv_print_debrief(), optimize, and repeat."

I think we should do the same thing here.

Contributor guide

Open the contributing guide

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 no files, tests, or specific scanning entry points. Start by locating the main scanning functions, then use profvis and debrief's pv_print_debrief() workflow to identify optimization targets; done would require demonstrated performance and memory improvements, but the target and success criteria are not specified.

Written by the indexing model from the issue text.

Assessment

Tech stack
r
Domain
performance
Issue type
Refactor
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.