verl/examples/grpo_trainer/run_qwen3_vl_8b_megatron.sh at main · aoshen02/verl · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
#!/usr/bin/env bash
# GRPO | vision | vLLM rollout | Megatron training | NVIDIA GPUs

set -xeuo pipefail

export CUDA_DEVICE_MAX_CONNECTIONS=1

# ---- user-adjustable ----
MODEL_PATH=${MODEL_PATH:-Qwen/Qwen3-VL-8B-Instruct}
NNODES=${NNODES:-1}
NGPUS_PER_NODE=${NGPUS_PER_NODE:-8}

train_batch_size=${TRAIN_BATCH_SIZE:-512}
ppo_mini_batch_size=${PPO_MINI_BATCH_SIZE:-128}
max_prompt_length=${MAX_PROMPT_LENGTH:-1024}
max_response_length=${MAX_RESPONSE_LENGTH:-2048}
ppo_max_token_len_per_gpu=${PPO_MAX_TOKEN_LEN_PER_GPU:-24576}

actor_lr=${ACTOR_LR:-1e-6}
kl_loss_coef=${KL_LOSS_COEF:-0.01}
entropy_coeff=${ENTROPY_COEFF:-0}

actor_tp=${ACTOR_TP:-2}
actor_pp=${ACTOR_PP:-2}

rollout_tp=${ROLLOUT_TP:-2}
rollout_gpu_mem_util=${ROLLOUT_GPU_MEM_UTIL:-0.6}
rollout_n=${ROLLOUT_N:-5}

total_epochs=${TOTAL_EPOCHS:-15}
save_freq=${SAVE_FREQ:-20}
test_freq=${TEST_FREQ:-5}

project_name=${PROJECT_NAME:-verl_grpo_geo3k}
experiment_name=${EXPERIMENT_NAME:-qwen3_vl_8b_vllm_megatron}
# ---- end user-adjustable ----
########################### parameter arrays ###########################

DATA=(
    algorithm.adv_estimator=grpo
    algorithm.use_kl_in_reward=False
    data.train_files=$HOME/data/geo3k/train.parquet
    data.val_files=$HOME/data/geo3k/test.parquet
    data.image_key=images
    data.train_batch_size=${train_batch_size}
    data.max_prompt_length=${max_prompt_length}
    data.max_response_length=${max_response_length}
    data.filter_overlong_prompts=True
    data.truncation='error'
)

MODEL=(
    actor_rollout_ref.model.path="$MODEL_PATH"
    actor_rollout_ref.model.use_remove_padding=True
    actor_rollout_ref.model.use_fused_kernels=True
)

ACTOR=(
    actor_rollout_ref.actor.optim.lr=${actor_lr}
    actor_rollout_ref.actor.ppo_mini_batch_size=${ppo_mini_batch_size}
    actor_rollout_ref.actor.use_dynamic_bsz=True
    actor_rollout_ref.actor.ppo_max_token_len_per_gpu=${ppo_max_token_len_per_gpu}
    actor_rollout_ref.actor.use_kl_loss=True
    actor_rollout_ref.actor.kl_loss_coef=${kl_loss_coef}
    actor_rollout_ref.actor.kl_loss_type=low_var_kl
    actor_rollout_ref.actor.entropy_coeff=${entropy_coeff}
    actor_rollout_ref.actor.megatron.tensor_model_parallel_size=${actor_tp}
    actor_rollout_ref.actor.megatron.pipeline_model_parallel_size=${actor_pp}
    actor_rollout_ref.actor.megatron.use_mbridge=True
)

ROLLOUT=(
    actor_rollout_ref.rollout.name=vllm
    actor_rollout_ref.rollout.tensor_model_parallel_size=${rollout_tp}
    actor_rollout_ref.rollout.gpu_memory_utilization=${rollout_gpu_mem_util}
    actor_rollout_ref.rollout.n=${rollout_n}
    actor_rollout_ref.rollout.log_prob_use_dynamic_bsz=True
    actor_rollout_ref.rollout.log_prob_max_token_len_per_gpu=${ppo_max_token_len_per_gpu}
)

REF=(
    actor_rollout_ref.ref.log_prob_use_dynamic_bsz=True
    actor_rollout_ref.ref.log_prob_max_token_len_per_gpu=${ppo_max_token_len_per_gpu}
    actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${actor_tp}
    actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${actor_pp}
    actor_rollout_ref.ref.megatron.use_mbridge=True
)

TRAINER=(
    trainer.balance_batch=True
    trainer.logger='["console","wandb"]'
    trainer.project_name=${project_name}
    trainer.experiment_name=${experiment_name}
    trainer.n_gpus_per_node=${NGPUS_PER_NODE}
    trainer.nnodes=${NNODES}
    trainer.save_freq=${save_freq}
    trainer.test_freq=${test_freq}
    trainer.total_epochs=${total_epochs}
)

EXTRA=(
    model_engine=megatron
)

########################### launch ###########################
python3 -m verl.trainer.main_ppo \
    "${DATA[@]}" \
    "${MODEL[@]}" \
    "${ACTOR[@]}" \
    "${ROLLOUT[@]}" \
    "${REF[@]}" \
    "${TRAINER[@]}" \
    "${EXTRA[@]}" \
    "$@"