Add configs and script to launch GRPO jobs to AWS cluster via Slurm

[daniellepintz]

Description

Add three configs for the FAIR AWS cluster, and a script to launch jobs.
Also make certain variables in the SlurmLauncher configurable.

Test Plan

Log onto AWS login node, and git clone torchforge:

git clone [email protected]:meta-pytorch/torchforge.git

Then setup your conda environment:

conda create -n forge python=3.12
conda activate forge
./scripts/install.sh

Run GRPO on Qwen3 8B model

./apps/grpo/slurm/submit.sh qwen3_8b

FAIR cluster job: https://www.internalfb.com/fair_hub/job/FAIR_SC/2362697/details
W&B: https://fairwandb.org/agentic-models/grpo-training/runs/yo82qdaq

Run GRPO on Qwen3 32B model

./apps/grpo/slurm/submit.sh qwen3_32b

FAIR cluster job: https://www.internalfb.com/fair_hub/job/FAIR_SC/2371720/details
W&B: https://fairwandb.org/agentic-models/grpo-training/runs/ryy2u75n

Run GRPO on Qwen3 30B A3B model

./apps/grpo/slurm/submit.sh qwen3_30b_a3b

FAIR cluster job: https://www.internalfb.com/fair_hub/job/FAIR_SC/2372099/details
W&B: https://fairwandb.org/agentic-models/grpo-training/runs/qmuo8whb

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.69%. Comparing base (7b8580a) to head (e0ba693).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #598      +/-   ##
==========================================
- Coverage   83.05%   82.69%   -0.37%     
==========================================
  Files          31       31              
  Lines        4036     3946      -90     
==========================================
- Hits         3352     3263      -89     
+ Misses        684      683       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[allenwang28]

lgtm, shall we remove the existing apps/grpo/qwen3_32b.yaml?

[daniellepintz]

yes, thanks will do in a separate PR!