File size: 3,728 Bytes
d4a7e84 | 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 | import os
import sys
import argparse
import subprocess
from tqdm import tqdm
# 添加当前目录到Python路径
sys.path.append(os.path.dirname(os.path.abspath(__file__)))
from eval.evaluation import simple_verify,majority_verify
def main():
parser = argparse.ArgumentParser(description="Sequence self-critique pipeline from previous solution in Majority.")
parser.add_argument('--model_name', type=str, required=True)
parser.add_argument('--max_completion_tokens', type=int, default=2048)
parser.add_argument('--dataset', type=str, required=True)
parser.add_argument('--method', type=str, default="Sequence")
parser.add_argument('--input_path', type=str, required=True)
# parser.add_argument('--output_dir', type=str, required=True)
parser.add_argument('--tensor_parallel_size', type=int, default=2)
parser.add_argument('--budget', type=int, default=8)
#parser.add_argument('--cuda_visible_devices', type=str, default=None)
args = parser.parse_args()
# 0. 检查输入文件是否存在
if not os.path.exists(args.input_path):
raise FileNotFoundError(f"Input file {args.input_path} not found")
# 1. 构建输出目录
result_dir = os.path.join(
'/home/tianqiu/tts_schedule/batch_infer/results',
args.dataset, args.model_name.replace('/', '_'), args.method
)
batch_dir = os.path.join(result_dir, "batch_data")
output_dir = os.path.join(result_dir, "output_data")
os.makedirs(result_dir, exist_ok=True)
os.makedirs(output_dir, exist_ok=True)
os.makedirs(batch_dir, exist_ok=True)
current_input_path = args.input_path
# 2. 数据准备
for i in range(args.budget):
# 修改 batch idx 为了连续
batch_dir_i = os.path.join(batch_dir, f"batch_{i}")
os.makedirs(batch_dir_i, exist_ok=True)
prepare_cmd = [
'python',
'/home/tianqiu/tts_schedule/batch_infer/src/sequence_data_prepare.py',
'--model_name', args.model_name,
'--max_completion_tokens', str(args.max_completion_tokens),
'--output_dir', batch_dir_i,
'--input_path', current_input_path,
'--budget', str(1)
]
print(f"[Pipeline] Running sequence data prepare: {' '.join(prepare_cmd)}")
subprocess.run(prepare_cmd, check=True)
# 3. Batch inference
batch_jsonl = os.path.join(batch_dir_i, f"batch_0.jsonl")
if not os.path.exists(batch_jsonl):
raise FileNotFoundError(f"Batch file {batch_jsonl} not found")
vllm_cmd = [
'python', '-m', 'vllm.entrypoints.openai.run_batch',
'-i', batch_jsonl,
'-o', os.path.join(batch_dir_i, f'output_0.jsonl'),
'--model', args.model_name,
'--tensor-parallel-size', str(args.tensor_parallel_size)
]
print(f"[Pipeline] Running batch inference: {' '.join(vllm_cmd)}")
env = os.environ.copy()
subprocess.run(vllm_cmd, check=True,env=env)
# 4. Extract
merge_cmd = [
'python', '/home/tianqiu/tts_schedule/batch_infer/src/output_extract.py',
'--input_dir', batch_dir_i,
'--extra_re',
'--dataset', args.dataset
]
print(f"[Pipeline] Running output merge: {' '.join(merge_cmd)}")
subprocess.run(merge_cmd, check=True)
current_input_path = os.path.join(batch_dir_i, f'parallel_merged_output.jsonl')
# evaluate
acc_path = os.path.join(batch_dir_i, f'acc.jsonl')
majority_verify(current_input_path,acc_path)
print(f"[Pipeline] All results saved in: {result_dir}")
# 合并
if __name__ == "__main__":
main()
|