MTSQL-R1/run_eval_text2sql.sh at main · taichengguo/MTSQL-R1 · 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
set -x

ulimit -n 65535
DATA_FOLDER="data/text2sql"
#DATA_FOLDER="data/gsm8k_our_test/"

# Dataset: Except Simple
# Except correct
PROJECT_DIR="$(pwd)"
CONFIG_PATH="$PROJECT_DIR/examples/sglang_multiturn/config"

#MODEL_PATH="Qwen/Qwen2.5-3B-Instruct"

#MODEL_PATH="Qwen/Qwen3-0.6B"
MODEL_PATH="Qwen/Qwen3-1.7B"
#MODEL_PATH="Qwen/Qwen3-4B"
#MODEL_PATH="Qwen/Qwen3-8B"
#MODEL_PATH="Qwen/Qwen3-14B"
#MODEL_PATH="Qwen/Qwen3-32B"


#MODEL_PATH="./text2sql_ckpt"


# Multi-turns multi-tools call
# 14B -> data;   1.7B RL
# Multiple times: Duplicate the training dataset
# 50 times
# CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
CUDA_VISIBLE_DEVICES=0,1,2,3 python3 -m verl.trainer.main_ppo \
--config-path="$CONFIG_PATH" \
--config-name='text2sql_multiturn_grpo' \
custom_reward_function.path=verl/utils/reward_score/text2sql.py  \
algorithm.adv_estimator=grpo \
data.train_files=$DATA_FOLDER/test.parquet \
data.val_files=$DATA_FOLDER/test.parquet  \
data.train_batch_size=64 \
data.max_prompt_length=7000 \
data.max_response_length=4096 \
data.filter_overlong_prompts=True \
data.truncation='error' \
data.return_raw_chat=True \
actor_rollout_ref.model.path=$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 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
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=8 \
actor_rollout_ref.rollout.tensor_model_parallel_size=1 \
actor_rollout_ref.rollout.name=sglang \
actor_rollout_ref.rollout.gpu_memory_utilization=0.8 \
actor_rollout_ref.rollout.n=1 \
actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=8 \
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='verl_grpo_text2sql' \
trainer.experiment_name="qwen3_eval_text2sqlf" \
trainer.val_before_train=True \
trainer.n_gpus_per_node=4 \
trainer.nnodes=1 \
trainer.save_freq=-1 \
trainer.test_freq=-1 \
actor_rollout_ref.rollout.multi_turn.tool_config_path="$PROJECT_DIR/examples/sglang_multiturn/config/tool_config/text2sql_tool_config.yaml" \
trainer.total_epochs=0 \
data.shuffle=False \
#> train_sql.log 2>&1
#data.val_batch_size=10 \
#    data.val_batch_size=50 \