feat: async performance recipe (under test)

[youngeunkwon0405]

What does this PR do ?

Wandb: https://wandb.ai/nvidia/async-grpo-perfscript-test

Test script

#!/bin/bash
huggingface-cli login --token $HF_TOKEN 

NUM_ACTOR_NODES=64
test_case="async-grpo-deepseek-v3-64n8g"
# test_case="async-grpo-llama3.1-8b-instruct-2n8g"
# test_case="async-grpo-qwen3-30ba3b-4n8g"
# test_case="async-grpo-qwen3-32b-8n8g"
# test_case="async-grpo-qwen3-235b-32n8g"


COMMAND="uv run ./examples/run_grpo_math.py \
--config examples/configs/recipes/llm/performance/${test_case}.yaml \
logger.wandb.project='async-grpo-perfscript-test'" \
CONTAINER=/lustre/fsw/portfolios/coreai/users/terryk/enroot-images/gitlab-master.nvidia.com/terryk/images/nemo-rl:main-2b55598e.squashfs \
HF_HOME=/lustre/fsw/portfolios/coreai/users/youngeunk/hf_home \
HF_DATASETS_CACHE=/lustre/fsw/portfolios/coreai/users/youngeunk/hf_home/cache \
NRL_FORCE_REBUILD_VENVS=true \
NCCL_NVLS_ENABLE=0 \
WANDB_API_KEY=$WANDB_API_KEY \
MOUNTS="/lustre:/lustre" \
sbatch \
    --nodes=${NUM_ACTOR_NODES} \
    --account=coreai_dlalgo_nemorl \
    --job-name=coreai_dlalgo_nemorl-async.test \
    --partition=batch \
    --time=01:00:00 \
    --gres=gpu:8 \
    ray.sub

Issues

List issues that this PR closes (syntax):

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• ...

[youngeunkwon0405]

Hi @guyueh1, I have run these tests, and it looks like the recipes are working fine.

For the DSV3, I used the local checkpoint in Yifu's directory and checked that it works.

I believe, the branch is now good to merge.