GRESO/train-scripts/math_qwen_7b_dm_greso.sh at main · Infini-AI-Lab/GRESO · 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
set -x

export VLLM_ATTENTION_BACKEND=XFORMERS

GPU_NUM=8
train_files="['./data/dapo_math/train.parquet', 'data/lighteval-math/train.parquet']"
test_files="['./data/math500/test.parquet', './data/amc/test.parquet', './data/aime2024/test.parquet', 'data/gaokao/test.parquet', 'data/minervamath/test.parquet', 'data/olympiadbench/test.parquet']"
project_name='greso'
mkdir -p data-log/$project_name
experiment_name='math_qwen_7b_dm_greso'

python -u -m verl.trainer.main_ppo \
    algorithm.adv_estimator=grpo \
    data.train_files="$train_files" \
    data.val_files="$test_files" \
    data.train_batch_size=32 \
    +data.dynamic_filtering=True \
    +data.dynamic_filtering_strategy=all_probabilistic \
    +data.p_easy=0.5 \
    +data.p_hard=0.5 \
    +data.target_zero_variance=0.25 \
    +data.sampling_batch_size=384 \
    +data.real_train_batch_size=256 \
    data.val_batch_size=8 \
    data.max_prompt_length=1536 \
    data.max_response_length=2560 \
    actor_rollout_ref.model.path=Qwen/Qwen2.5-Math-7B \
    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=4 \
    +ttis.type=dynamic_sampling \
    actor_rollout_ref.actor.use_kl_loss=False \
    actor_rollout_ref.actor.kl_loss_coef=0.00 \
    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
    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=vllm \
    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
    actor_rollout_ref.rollout.n=8 \
    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.val_kwargs.n=4 \
    actor_rollout_ref.rollout.val_kwargs.do_sample=True \
    actor_rollout_ref.rollout.val_kwargs.top_p=0.7 \
    actor_rollout_ref.rollout.val_kwargs.temperature=1 \
    algorithm.kl_ctrl.kl_coef=0.000 \
    trainer.critic_warmup=0 \
    trainer.logger=['console','wandb'] \
    trainer.project_name=$project_name \
    trainer.experiment_name=$experiment_name \
    trainer.n_gpus_per_node=$GPU_NUM \
    trainer.nnodes=1 \
    trainer.save_freq=40 \
    trainer.test_freq=50 \
    +trainer.val_before_train=False \
    +trainer.max_steps=1001 \
    trainer.total_epochs=2000 \
    | tee data-log/$project_name/$experiment_name.log