unify the filename and training menthods of the two scripts

Author: tardis-key

examples/grpo_trainer/run_qwen3_32b_npu.sh renamed to examples/grpo_trainer/run_qwen3-32b_npu.sh

@@ -1,16 +1,24 @@
 set -x
 
+project_name='GRPO-Qwen3'
+exp_name='GRPO-Qwen3-32b-npu'
+gen_tp=4
+RAY_DATA_HOME=${RAY_DATA_HOME:-"${HOME}/verl"}
+MODEL_PATH=${MODEL_PATH:-"${RAY_DATA_HOME}/models/Qwen3-32B"}
+TRAIN_FILE=${TRAIN_FILE:-"${RAY_DATA_HOME}/data/gsm8k/train.parquet"}
+TEST_FILE=${TEST_FILE:-"${RAY_DATA_HOME}/data/gsm8k/test.parquet"}
+
 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_files="${TRAIN_FILE}" \
+    data.val_files="${TEST_FILE}" \
     data.train_batch_size=1024 \
     data.max_prompt_length=2048 \
     data.max_response_length=2048 \
     data.filter_overlong_prompts=True \
     data.truncation='error' \
     data.shuffle=False \
-    actor_rollout_ref.model.path=Qwen/Qwen3-32B \
+    actor_rollout_ref.model.path=${MODEL_PATH} \
     actor_rollout_ref.actor.optim.lr=1e-6 \
     actor_rollout_ref.model.use_remove_padding=True \
     actor_rollout_ref.actor.ulysses_sequence_parallel_size=4 \
@@ -27,7 +35,7 @@ python3 -m verl.trainer.main_ppo \
     actor_rollout_ref.actor.fsdp_config.param_offload=True \
     actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
     actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=8 \
-    actor_rollout_ref.rollout.tensor_model_parallel_size=4 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=${gen_tp} \
     actor_rollout_ref.rollout.name=vllm \
     actor_rollout_ref.rollout.gpu_memory_utilization=0.7 \
     actor_rollout_ref.rollout.n=4 \
@@ -40,8 +48,8 @@ python3 -m verl.trainer.main_ppo \
     algorithm.use_kl_in_reward=False \
     trainer.critic_warmup=0 \
     trainer.logger=['console','tensorboard'] \
-    trainer.project_name='verl_grpo_example_gsm8k_fsdp' \
-    trainer.experiment_name='qwen3_32b_fsdp' \
+    trainer.project_name="${project_name}" \
+    trainer.experiment_name="${exp_name}" \
     trainer.n_gpus_per_node=16 \
     trainer.nnodes=2 \
     trainer.resume_from_path=checkpoints/ \

examples/grpo_trainer/run_qwen3-8b_npu.sh

@@ -1,23 +1,15 @@
-# Tested successfully on the hiyouga/verl:ngc-th2.6.0-cu126-vllm0.8.4-flashinfer0.2.2-cxx11abi0 image.
-# It outperforms the Qwen2 7B base model by two percentage points on the test set of GSM8K.
 set -x
 
-project_name='GRPO'
+project_name='GRPO-Qwen3'
 exp_name='GRPO-Qwen3-8B-npu'
 gen_tp=2
-WORKING_DIR=${WORKING_DIR:-"./"}
-RUNTIME_ENV=${RUNTIME_ENV:-"${WORKING_DIR}/verl/trainer/runtime_env.yaml"}
-RAY_ADDRESS=${RAY_ADDRESS:-"http://localhost:8265"}
 RAY_DATA_HOME=${RAY_DATA_HOME:-"${HOME}/verl"}
 MODEL_PATH=${MODEL_PATH:-"${RAY_DATA_HOME}/models/Qwen3-8B"}
 CKPTS_DIR=${CKPTS_DIR:-"${RAY_DATA_HOME}/ckpts/${project_name}/${exp_name}"}
 TRAIN_FILE=${TRAIN_FILE:-"${RAY_DATA_HOME}/data/dapo-math-17k.parquet"}
 TEST_FILE=${TEST_FILE:-"${RAY_DATA_HOME}/data/aime-2024.parquet"}
 
-ray job submit --no-wait --runtime-env="${RUNTIME_ENV}" \
-    --working-dir "${WORKING_DIR}" \
-    --address "${RAY_ADDRESS}" \
-    -- python3 -m verl.trainer.main_ppo \
+python3 -m verl.trainer.main_ppo \
     algorithm.adv_estimator=grpo \
     data.train_files="${TRAIN_FILE}" \
     data.val_files="${TEST_FILE}" \
@@ -26,7 +18,7 @@ ray job submit --no-wait --runtime-env="${RUNTIME_ENV}" \
     data.max_response_length=1024 \
     data.filter_overlong_prompts=True \
     data.truncation='error' \
-    actor_rollout_ref.model.path=${MODEL_PATH}} \
+    actor_rollout_ref.model.path=${MODEL_PATH} \
     actor_rollout_ref.actor.optim.lr=1e-6 \
     actor_rollout_ref.model.use_remove_padding=True \
     actor_rollout_ref.actor.ppo_mini_batch_size=64 \