[sglang] feat: add native sgl server

examples/sglang_multiturn/config/gsm8k_multiturn_grpo_server.yaml

@@ -0,0 +1,28 @@
+hydra:
+  searchpath:
+    - file://verl/trainer/config
+
+defaults:
+  - ppo_trainer
+  - _self_
+
+data:
+  max_prompt_length: 1024
+  max_response_length: 1024
+  train_batch_size: 256
+  return_raw_chat: True
+
+actor_rollout_ref:
+  hybrid_engine: True
+  rollout:
+    name: sglang
+    multi_turn:
+      enable: True
+      max_assistant_turns: 5
+    sglang_rollout_mode: server
+    server:
+      timeout: 60
+      max_attempts: 3
+      retry_delay: 2
+      max_connections: 1000
+      max_start_wait_time: 300.0

examples/sglang_multiturn/run_qwen2.5-3b_gsm8k_multiturn_4xgpu_server.sh

@@ -0,0 +1,60 @@
+# run on 4xH100
+# make sure your current working directory is the root of the project
+
+set -x
+export HYDRA_FULL_ERROR=1
+ulimit -n 65535
+
+PROJECT_DIR="$(pwd)"
+CONFIG_PATH="$PROJECT_DIR/examples/sglang_multiturn/config"
+
+python3 -m verl.trainer.main_ppo \
+    --config-path="$CONFIG_PATH" \
+    --config-name='gsm8k_multiturn_grpo_server' \
+    algorithm.adv_estimator=grpo \
+    data.train_batch_size=256 \
+    data.max_prompt_length=1024 \
+    data.max_response_length=1024 \
+    data.filter_overlong_prompts=True \
+    data.truncation='error' \
+    data.return_raw_chat=True \
+    actor_rollout_ref.model.path=Qwen/Qwen2.5-3B-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=32 \
+    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.actor.entropy_coeff=0 \
+    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=32 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=4 \
+    actor_rollout_ref.rollout.name=sglang \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=16 \
+    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32 \
+    actor_rollout_ref.ref.fsdp_config.param_offload=True \
+    algorithm.use_kl_in_reward=False \
+    trainer.critic_warmup=0 \
+    trainer.logger='["console", "wandb"]' \
+    trainer.project_name='gsm8k_async_rl_server' \
+    trainer.experiment_name='qwen2.5-3b_function_rm-gsm8k-async-sgl-multi-w-tool-verify-n16-4cards' \
+    trainer.n_gpus_per_node=4 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=20 \
+    trainer.total_epochs=15 \
+    actor_rollout_ref.actor.ppo_max_token_len_per_gpu=8192 \
+    actor_rollout_ref.rollout.log_prob_max_token_len_per_gpu=8192 \
+    actor_rollout_ref.ref.log_prob_max_token_len_per_gpu=8192 \
+    critic.ppo_max_token_len_per_gpu=8192 \
+    critic.forward_max_token_len_per_gpu=8192 \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    actor_rollout_ref.rollout.multi_turn.tool_config_path="$PROJECT_DIR/examples/sglang_multiturn/config/tool_config/gsm8k_tool_config.yaml" \
+    actor_rollout_ref.rollout.multi_turn.interaction_config_path="$PROJECT_DIR/examples/sglang_multiturn/config/interaction_config/gsm8k_interaction_config.yaml" \
+    actor_rollout_ref.rollout.multi_turn.max_user_turns=1 \
+    $@

examples/sglang_multiturn/run_qwen2.5-3b_gsm8k_multiturn_server.sh

@@ -0,0 +1,63 @@
+# run on 8xH100
+# make sure your current working directory is the root of the project
+
+set -x
+
+ulimit -n 65535
+
+PROJECT_DIR="$(pwd)"
+CONFIG_PATH="$PROJECT_DIR/examples/sglang_multiturn/config"
+
+function now() {
+    date '+%d-%H-%M'
+}
+
+EXPERIMENT_NAME="qwen2.5-3b_baseline_$(now)"
+
+python3 -m verl.trainer.main_ppo \
+    --config-path="$CONFIG_PATH" \
+    --config-name='gsm8k_multiturn_grpo_server' \
+    algorithm.adv_estimator=grpo \
+    data.train_batch_size=256 \
+    data.max_prompt_length=1024 \
+    data.max_response_length=1024 \
+    data.filter_overlong_prompts=True \
+    data.truncation='error' \
+    data.return_raw_chat=True \
+    actor_rollout_ref.model.path=Qwen/Qwen2.5-3B-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=32 \
+    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.actor.entropy_coeff=0 \
+    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=32 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=1 \
+    actor_rollout_ref.rollout.name=sglang \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.85 \
+    actor_rollout_ref.rollout.multi_stage_wake_up=True \
+    actor_rollout_ref.rollout.n=16 \
+    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32 \
+    actor_rollout_ref.ref.fsdp_config.param_offload=True \
+    actor_rollout_ref.rollout.over_sample_rate=0 \
+    algorithm.use_kl_in_reward=False \
+    trainer.critic_warmup=0 \
+    trainer.logger='["console","wandb"]' \
+    trainer.project_name='multi-turn-grpo-qwen2.5-3b-sglang' \
+    trainer.experiment_name=$EXPERIMENT_NAME \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=20 \
+    trainer.val_before_train=True \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    actor_rollout_ref.rollout.multi_turn.tool_config_path="$PROJECT_DIR/examples/sglang_multiturn/config/tool_config/gsm8k_tool_config.yaml" \
+    trainer.total_epochs=15 \
+    actor_rollout_ref.rollout.update_weights_bucket_megabytes=512 $@
+