EasyR1/examples/qwen2.5_bio.sh

#!/bin/bash

set -x

export PYTHONUNBUFFERED=1
export WANDB_BASE_URL=https://api.bandw.top

MODEL_PATH=/oss/wangyujia/BIO/pretrain_output/qwen2.5-7b-instruct-bio/bio_all/save1epoch/checkpoint-1300  # replace it with your local file path

python3 -m verl.trainer.main \
    config=config.yaml \
    data.train_files=/nas/shared/kilab/wangyujia/rl_data/deeplocmulti@train \
    data.val_files=/nas/shared/kilab/wangyujia/rl_data/deeplocmulti@validation \
    data.max_prompt_length=4096 \
    data.max_response_length=16384\
    worker.actor.model.model_path=${MODEL_PATH} \
    worker.rollout.tensor_parallel_size=1 \
    worker.rollout.n=5 \
    trainer.experiment_name=qwen2.5_7b_bio_06182042 \
    trainer.n_gpus_per_node=8 \
    trainer.total_epochs=1 \
    trainer.save_checkpoint_path=/oss/wangyujia/BIO/rl/qwen2.5_7b_bio_06182042