facebookresearch / facebookresearch/fairo

[HiTL] Spin up more runs with different random seed in nsp model retrain infra

Open
#892 0 comments 0 reactions 1 assignee Claimed by @ethan-carlson View on GitHub
Dominant language
Jupyter Notebook
Stars
929
Forks
123
PR merge metrics
No merged PRs in 30d

Description

## Type of Issue

Select the type of issue:
- [ ] Bug report (to report a bug)
- [x] Feature request (to request an additional feature)
- [ ] Tracker (I am just using this as a tracker)
- [ ] Refactor request
- [ ] Documentation Ask

## Description

Currently we spin up 6 runs with different parameters in model sweep. Those are with same random seed (batch_id).

As discussed, we should expand sweep size -- use N different random seeds, each with 6 runs (the parameter config file should be the same for those N random seeds). The total number of runs in one sweep would be N * 6.

Note that we should use the same mask splits for all the runs in one sweep. That is, keep using batch_id as random seed while generating mask splits (or load & store), then change random seeds to spin up sweep runs.

## Checklist

Use this checklist if you are using this issue as a tracker:

- [ ] Task 1
- [ ] Task 2
- [ ] Task 3 ...

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.