Add GRPO CI to FSDP and Megatron simple e2e. (#711)

For longer tests, may check `example/grpo_trainer` folder, these 2
backends can align within 200 steps, but for more steps, megatron seems
not able to reach loss convergence.

TODO: Extended testing over longer time ranges is required to further
validate.

.github/workflows/e2e_grpo.yml

+name: e2e_grpo
+
+on:
+  # Trigger the workflow on push or pull request,
+  # but only for the main branch
+  push:
+    branches:
+      - main
+      - v0.2.x
+    paths:
+      - "**/*.py"
+      - .github/workflows/e2e_grpo.yml
+  pull_request:
+    branches:
+      - main
+      - v0.2.x
+    paths:
+      - "**/*.py"
+      - "verl/trainer/config/*.yaml"
+      - .github/workflows/e2e_grpo.yml
+      - "tests/e2e/*.sh"
+
+# Cancel jobs on the same ref if a new one is triggered
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
+# Declare permissions just read content.
+permissions: 
+  contents: read
+
+jobs:
+  e2e_gsm8k_megatron:
+    runs-on: [self-hosted, l20-0]
+    timeout-minutes: 60 # Increase this timeout value as needed
+    env:
+      HTTP_PROXY: ${{ secrets.PROXY_HTTP }}
+      HTTPS_PROXY: ${{ secrets.PROXY_HTTPS }}
+      NO_PROXY: "localhost,127.0.0.1"
+      HF_HUB_ENABLE_HF_TRANSFER: 1
+    container:
+      image: whatcanyousee/verl:vemlp-th2.4.0-cu124-vllm0.6.3-ray2.10-te2.0-megatron0.11.0-v0.0.6
+      options: --gpus all --shm-size=10g
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+            fetch-depth: 0
+      - name: Install the current repository
+        run: |
+          pip3 install hf_transfer
+          pip3 install -e .[test]
+      - name: Prepare gsm8k dataset
+        run: |
+          python3 examples/data_preprocess/gsm8k.py
+      - name: Running GRPO gsm8k e2e training tests with FSDP on 8 L20 GPUs (Deepseek)
+        run: |
+          ray stop --force
+          bash tests/e2e/run_deepseek_grpo.sh
+      - name: Running GRPO gsm8k e2e training tests with 3D parallelism on 8 L20 GPUs with Megatron (Deepseek)
+        run: |
+          ray stop --force
+          bash tests/e2e/run_deepseek_grpo_megatron.sh
+      - name: Running GRPO gsm8k e2e training tests with FSDP on 8 L20 GPUs (Qwen)
+        run: |
+          ray stop --force
+          bash tests/e2e/run_qwen_grpo.sh
+      - name: Running GRPO gsm8k e2e training tests with 3D parallelism on 8 L20 GPUs with Megatron (Qwen)
+        run: |
+          ray stop --force
+          bash tests/e2e/run_qwen_grpo_megatron.sh
\ No newline at end of file

.github/workflows/e2e_gsm8k_megatron.yml

@@ -40,7 +40,7 @@ jobs:
       NO_PROXY: "localhost,127.0.0.1"
       HF_HUB_ENABLE_HF_TRANSFER: 1
     container:
-      image: whatcanyousee/verl:vemlp-th2.4.0-cu124-vllm0.6.3-ray2.10-te2.0-megatron0.11.0-v0.0.5
+      image: whatcanyousee/verl:vemlp-th2.4.0-cu124-vllm0.6.3-ray2.10-te2.0-megatron0.11.0-v0.0.6
       options: --gpus all --shm-size=10g
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
@@ -56,10 +56,8 @@ jobs:
       - name: Running gsm8k e2e training tests with 3D parallelism on 8 L20 GPUs with Megatron (Deepseek)
         run: |
           ray stop --force
-          export PYTHONPATH=$PYTHONPATH:/opt/nvidia/Megatron-LM
           bash tests/e2e/run_deepseek_megatron_parallelism.sh
       - name: Running gsm8k e2e training tests with 3D parallelism on 8 L20 GPUs with Megatron (Qwen)
         run: |
           ray stop --force
-          export PYTHONPATH=$PYTHONPATH:/opt/nvidia/Megatron-LM
           bash tests/e2e/run_qwen_megatron_parallelism.sh
\ No newline at end of file

tests/e2e/run_deepseek_grpo.sh

+set -x
+
+export VLLM_ATTENTION_BACKEND=XFORMERS
+
+python3 -m verl.trainer.main_ppo \
+    algorithm.adv_estimator=grpo \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=1024 \
+    data.max_prompt_length=512 \
+    data.max_response_length=1024 \
+    data.filter_overlong_prompts=True \
+    data.truncation='error' \
+    actor_rollout_ref.model.path=deepseek-ai/deepseek-coder-1.3b-instruct \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.model.use_remove_padding=True \
+    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
+    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=80 \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.actor.fsdp_config.param_offload=False \
+    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=160 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=5 \
+    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=160 \
+    actor_rollout_ref.ref.fsdp_config.param_offload=True \
+    algorithm.kl_ctrl.kl_coef=0.001 \
+    trainer.critic_warmup=0 \
+    trainer.logger=['console'] \
+    trainer.project_name='verl_grpo_example_gsm8k' \
+    trainer.experiment_name='deepseek_llm_7b_function_rm' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=15 \
+    trainer.total_training_steps=2 $@
\ No newline at end of file

tests/e2e/run_deepseek_grpo_megatron.sh

+set -x
+
+export VLLM_ATTENTION_BACKEND=XFORMERS
+
+python3 -m verl.trainer.main_ppo --config-path=config \
+    --config-name='ppo_megatron_trainer.yaml'\
+    algorithm.adv_estimator=grpo \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=1024 \
+    data.max_prompt_length=1024 \
+    data.max_response_length=1024 \
+    data.filter_overlong_prompts=True \
+    data.truncation='error' \
+    actor_rollout_ref.model.path=deepseek-ai/deepseek-coder-1.3b-instruct \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
+    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
+    actor_rollout_ref.actor.megatron.pipeline_model_parallel_size=2 \
+    actor_rollout_ref.actor.megatron.virtual_pipeline_model_parallel_size=2 \
+    actor_rollout_ref.actor.megatron.tensor_model_parallel_size=2 \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=4 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=5 \
+    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=4 \
+    algorithm.kl_ctrl.kl_coef=0.001 \
+    trainer.critic_warmup=0 \
+    trainer.logger=['console'] \
+    trainer.project_name='verl_grpo_example_gsm8k' \
+    trainer.experiment_name='deepseek_llm_7b_function_rm_math_megatron' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=15 \
+    trainer.total_training_steps=2 $@
\ No newline at end of file

tests/e2e/run_qwen_grpo.sh

+set -x
+
+export VLLM_ATTENTION_BACKEND=XFORMERS
+
+python3 -m verl.trainer.main_ppo \
+    algorithm.adv_estimator=grpo \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=1024 \
+    data.max_prompt_length=512 \
+    data.max_response_length=1024 \
+    data.filter_overlong_prompts=True \
+    data.truncation='error' \
+    actor_rollout_ref.model.path=Qwen/Qwen2.5-0.5B \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.model.use_remove_padding=True \
+    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
+    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16 \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.actor.fsdp_config.param_offload=True \
+    actor_rollout_ref.actor.fsdp_config.optimizer_offload=True \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=5 \
+    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16 \
+    actor_rollout_ref.ref.fsdp_config.param_offload=True \
+    algorithm.kl_ctrl.kl_coef=0.001 \
+    trainer.critic_warmup=0 \
+    trainer.logger=['console'] \
+    trainer.project_name='verl_grpo_example_gsm8k' \
+    trainer.experiment_name='qwen2_7b_function_rm' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=15 \
+    trainer.total_training_steps=2 $@
\ No newline at end of file

tests/e2e/run_qwen_grpo_megatron.sh

+set -x
+
+export VLLM_ATTENTION_BACKEND=XFORMERS
+
+python3 -m verl.trainer.main_ppo --config-path=config \
+    --config-name='ppo_megatron_trainer.yaml'\
+    algorithm.adv_estimator=grpo \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=1024 \
+    data.max_prompt_length=512 \
+    data.max_response_length=1024 \
+    data.filter_overlong_prompts=True \
+    data.truncation='error' \
+    actor_rollout_ref.model.path=Qwen/Qwen2.5-0.5B \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
+    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
+    actor_rollout_ref.actor.megatron.pipeline_model_parallel_size=2 \
+    actor_rollout_ref.actor.megatron.virtual_pipeline_model_parallel_size=2 \
+    actor_rollout_ref.actor.megatron.tensor_model_parallel_size=2 \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=4 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=5 \
+    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=4 \
+    algorithm.kl_ctrl.kl_coef=0.001 \
+    trainer.critic_warmup=0 \
+    trainer.logger=['console'] \
+    trainer.project_name='verl_grpo_example_gsm8k' \
+    trainer.experiment_name='qwen2_7b_function_rm_megatron' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=15 \
+    trainer.total_training_steps=2 $@
\ No newline at end of file