THUDM / THUDM/slime

Brainstorms about visualization and post-hoc analysis

Open
#85 3 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
8.5k
Forks
1.3k
Avg merge
5h 36m
Merged PRs (30d)
22

Description

Since RL is hard to debug, I am wondering whether make everything visualizable would be somehow helpful. For example, a most naive approach is:

  • Dump all data to disk, including rollout data and training logprob etc
    • Currently we already have a "save data for debug replay" feature, which may be combined
  • Create a visualizer to visualize the data, e.g. clickable tables to look at each rollout sample, their token-level prob and ref logprob and whatever, and also some plotly charts for statistics etc

EDIT: potential prerequisite: the buffer refactor in https://github.com/THUDM/slime/issues/80

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

Start by reviewing the existing “save data for debug replay” feature and the buffer refactor tracked in issue #80. The issue does not name files, entry points, tests, or a defined completion criterion; work would first require narrowing the visualization and post-hoc analysis scope.

Written by the indexing model from the issue text.

Assessment

Tech stack
plotly, python
Domain
data-visualization, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.