[misc] minor update

examples/baselines/qwen2_5_vl_3b_clevr.sh

@@ -2,7 +2,7 @@ set -x
 
 MODEL_PATH=Qwen/Qwen2.5-VL-3B-Instruct  # replace it with your local file path
 
-SYSTEM_PROMPT="""A conversation between User and Assistant. The user asks a question, and the Assistant solves it. The assistant
+FORMAT_PROMPT="""A conversation between User and Assistant. The user asks a question, and the Assistant solves it. The assistant
  first thinks about the reasoning process in the mind and then provides the user with the answer. The reasoning
  process and answer are enclosed within <think> </think> and <answer> </answer> tags, respectively, i.e.,
  <think> reasoning process here </think><answer> answer here </answer>"""
@@ -11,10 +11,9 @@ python3 -m verl.trainer.main \
     config=examples/config.yaml \
     data.train_files=BUAADreamer/clevr_count_70k@train \
     data.val_files=BUAADreamer/clevr_count_70k@test \
-    data.system_prompt="${SYSTEM_PROMPT}" \
+    data.format_prompt="${FORMAT_PROMPT}" \
     worker.actor.model.model_path=${MODEL_PATH} \
     worker.rollout.tensor_parallel_size=1 \
-    worker.rollout.enable_chunked_prefill=false \
-    worker.reward.compute_score=r1v \
+    worker.reward.score_function=r1v \
     trainer.experiment_name=qwen2_5_vl_3b_clevr \
     trainer.n_gpus_per_node=2

examples/baselines/qwen2_5_vl_3b_geoqa8k.sh

@@ -2,7 +2,7 @@ set -x
 
 MODEL_PATH=Qwen/Qwen2.5-VL-3B-Instruct  # replace it with your local file path
 
-SYSTEM_PROMPT="""A conversation between User and Assistant. The user asks a question, and the Assistant solves it. The assistant
+FORMAT_PROMPT="""A conversation between User and Assistant. The user asks a question, and the Assistant solves it. The assistant
  first thinks about the reasoning process in the mind and then provides the user with the answer. The reasoning
  process and answer are enclosed within <think> </think> and <answer> </answer> tags, respectively, i.e.,
  <think> reasoning process here </think><answer> answer here </answer>"""
@@ -11,10 +11,9 @@ python3 -m verl.trainer.main \
     config=examples/config.yaml \
     data.train_files=leonardPKU/GEOQA_8K_R1V@train \
     data.val_files=leonardPKU/GEOQA_8K_R1V@test \
-    data.system_prompt="${SYSTEM_PROMPT}" \
+    data.format_prompt="${FORMAT_PROMPT}" \
     worker.actor.model.model_path=${MODEL_PATH} \
     worker.rollout.tensor_parallel_size=1 \
-    worker.rollout.enable_chunked_prefill=false \
-    worker.reward.compute_score=r1v \
+    worker.reward.score_function=r1v \
     trainer.experiment_name=qwen2_5_vl_3b_geoqa8k \
     trainer.n_gpus_per_node=8

examples/config.yaml

@@ -68,7 +68,8 @@ worker:
 
   reward:
     reward_type: function
-    compute_score: math
+    score_function: math
+    skip_special_tokens: true
 
 trainer:
   total_episodes: 15
@@ -80,7 +81,7 @@ trainer:
   val_freq: 5  # -1 to disable
   val_before_train: true
   val_only: false
-  val_generations_to_log: 1
+  val_generations_to_log: 3
   save_freq: 5  # -1 to disable
   save_limit: 3  # -1 to disable
   save_checkpoint_path: null

examples/qwen2_5_7b_math_grpo.sh

@@ -2,14 +2,14 @@ set -x
 
 MODEL_PATH=Qwen/Qwen2.5-7B-Instruct  # replace it with your local file path
 
-SYSTEM_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
+FORMAT_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
  The reasoning process MUST BE enclosed within <think> </think> tags. The final answer MUST BE put in \boxed{}."""
 
 python3 -m verl.trainer.main \
     config=examples/config.yaml \
     data.train_files=hiyouga/math12k@train \
     data.val_files=hiyouga/math12k@test \
-    data.system_prompt="${SYSTEM_PROMPT}" \
+    data.format_prompt="${FORMAT_PROMPT}" \
     worker.actor.model.model_path=${MODEL_PATH} \
     trainer.experiment_name=qwen2_5_7b_math_grpo \
     trainer.n_gpus_per_node=8

examples/qwen2_5_vl_32b_geo3k_grpo.sh

@@ -2,20 +2,19 @@ set -x
 
 MODEL_PATH=Qwen/Qwen2.5-VL-32B-Instruct  # replace it with your local file path
 
-SYSTEM_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
+FORMAT_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
  The reasoning process MUST BE enclosed within <think> </think> tags. The final answer MUST BE put in \boxed{}."""
 
 python3 -m verl.trainer.main \
     config=examples/config.yaml \
     data.train_files=hiyouga/geometry3k@train \
     data.val_files=hiyouga/geometry3k@test \
-    data.system_prompt="${SYSTEM_PROMPT}" \
+    data.format_prompt="${FORMAT_PROMPT}" \
     worker.actor.model.model_path=${MODEL_PATH} \
     worker.actor.micro_batch_size_per_device_for_update=1 \
     worker.actor.micro_batch_size_per_device_for_experience=8 \
     worker.actor.fsdp.torch_dtype=bf16 \
     worker.actor.optim.strategy=adamw_bf16 \
     worker.rollout.tensor_parallel_size=8 \
-    worker.rollout.enable_chunked_prefill=false \
     trainer.experiment_name=qwen2_5_vl_32b_geo_grpo \
     trainer.n_gpus_per_node=8

examples/qwen2_5_vl_3b_geo3k_grpo.sh

@@ -2,16 +2,15 @@ set -x
 
 MODEL_PATH=Qwen/Qwen2.5-VL-3B-Instruct  # replace it with your local file path
 
-SYSTEM_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
+FORMAT_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
  The reasoning process MUST BE enclosed within <think> </think> tags. The final answer MUST BE put in \boxed{}."""
 
 python3 -m verl.trainer.main \
     config=examples/config.yaml \
     data.train_files=hiyouga/geometry3k@train \
     data.val_files=hiyouga/geometry3k@test \
-    data.system_prompt="${SYSTEM_PROMPT}" \
+    data.format_prompt="${FORMAT_PROMPT}" \
     worker.actor.model.model_path=${MODEL_PATH} \
     worker.rollout.tensor_parallel_size=1 \
-    worker.rollout.enable_chunked_prefill=false \
     trainer.experiment_name=qwen2_5_vl_3b_geo_grpo \
     trainer.n_gpus_per_node=2

examples/qwen2_5_vl_7b_geo3k_grpo.sh

@@ -2,15 +2,14 @@ set -x
 
 MODEL_PATH=Qwen/Qwen2.5-VL-7B-Instruct  # replace it with your local file path
 
-SYSTEM_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
+FORMAT_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
  The reasoning process MUST BE enclosed within <think> </think> tags. The final answer MUST BE put in \boxed{}."""
 
 python3 -m verl.trainer.main \
     config=examples/config.yaml \
     data.train_files=hiyouga/geometry3k@train \
     data.val_files=hiyouga/geometry3k@test \
-    data.system_prompt="${SYSTEM_PROMPT}" \
+    data.format_prompt="${FORMAT_PROMPT}" \
     worker.actor.model.model_path=${MODEL_PATH} \
-    worker.rollout.enable_chunked_prefill=false \
     trainer.experiment_name=qwen2_5_vl_7b_geo_grpo \
     trainer.n_gpus_per_node=8

examples/qwen2_5_vl_7b_geo3k_reinforce.sh

@@ -2,16 +2,15 @@ set -x
 
 MODEL_PATH=Qwen/Qwen2.5-VL-7B-Instruct  # replace it with your local file path
 
-SYSTEM_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
+FORMAT_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
  The reasoning process MUST BE enclosed within <think> </think> tags. The final answer MUST BE put in \boxed{}."""
 
 python3 -m verl.trainer.main \
     config=examples/config.yaml \
     data.train_files=hiyouga/geometry3k@train \
     data.val_files=hiyouga/geometry3k@test \
-    data.system_prompt="${SYSTEM_PROMPT}" \
+    data.format_prompt="${FORMAT_PROMPT}" \
     worker.actor.model.model_path=${MODEL_PATH} \
-    worker.rollout.enable_chunked_prefill=false \
     algorithm.adv_estimator=reinforce_plus_plus \
     trainer.experiment_name=qwen2_5_vl_7b_geo_reinforce_pp \
     trainer.n_gpus_per_node=8

examples/qwen2_5_vl_7b_geo3k_swanlab.sh

@@ -2,16 +2,15 @@ set -x
 
 MODEL_PATH=Qwen/Qwen2.5-VL-7B-Instruct  # replace it with your local file path
 
-SYSTEM_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
+FORMAT_PROMPT="""You FIRST think about the reasoning process as an internal monologue and then provide the final answer.
  The reasoning process MUST BE enclosed within <think> </think> tags. The final answer MUST BE put in \boxed{}."""
 
 python3 -m verl.trainer.main \
     config=examples/config.yaml \
     data.train_files=hiyouga/geometry3k@train \
     data.val_files=hiyouga/geometry3k@test \
-    data.system_prompt="${SYSTEM_PROMPT}" \
+    data.format_prompt="${FORMAT_PROMPT}" \
     worker.actor.model.model_path=${MODEL_PATH} \
-    worker.rollout.enable_chunked_prefill=false \
     trainer.experiment_name=qwen2_5_vl_7b_geo_grpo \
     trainer.logger=['console','swanlab'] \
     trainer.n_gpus_per_node=8