Files
2024-09-25 18:04:29 +08:00

75 lines
2.5 KiB
Python
Executable File

import gc
import json
import os
from argparse import ArgumentParser
from pathlib import Path
import torch
from humaneval_qwen2_base_multilangs import HumanEval as MultiPLERunner
from vllm import LLM
from vllm.distributed.parallel_state import destroy_distributed_environment, destroy_model_parallel
DATA_ROOT = str(Path(__file__).joinpath("../data").resolve())
print(f"{DATA_ROOT = }")
if __name__ == "__main__":
parser = ArgumentParser()
parser.add_argument("--logdir", type=str, default="")
parser.add_argument("--modelpath", type=str, default="")
parser.add_argument("--tp", default=1, type=int)
parser.add_argument("--just_eval", action="store_true", default=False)
parser.add_argument("--langs", type=str, nargs="+", default=["java", "cpp", "js", "cs", "php", "sh", "ts"])
parser.add_argument("--no_batching", action="store_true", default=False)
parser.add_argument("--no_new_line_at_last", action="store_true", default=False)
args = parser.parse_args()
logdir = args.logdir
if logdir == "":
logdir = "debug/"
Path(logdir).mkdir(exist_ok=True, parents=True)
evaluator = MultiPLERunner(
data_root=DATA_ROOT,
max_seq_len=4096,
log_dir=logdir,
n_sample=1,
max_gen_len=500,
no_batching=args.no_batching,
no_new_line_at_last=args.no_new_line_at_last,
)
if not args.just_eval:
print(f"TP = {args.tp = }")
model = LLM(
model=args.modelpath,
max_model_len=8192,
tensor_parallel_size=args.tp,
gpu_memory_utilization=0.95,
enforce_eager=True,
disable_custom_all_reduce=True,
trust_remote_code=True,
distributed_executor_backend="ray", # temp fix
)
else:
model = None
print(f"Skip model loading, just eval...")
os.environ["TOKENIZERS_PARALLELISM"] = "false"
aggr_result = evaluator.eval_model_multiple(model, langs=args.langs, just_eval=args.just_eval)
res_file = Path(logdir).joinpath("results.json")
with Path(res_file).open("w") as f:
json.dump(aggr_result, f, ensure_ascii=False, indent=2)
if not args.just_eval:
print(f"Try cleanup...")
destroy_model_parallel()
destroy_distributed_environment()
del model.llm_engine.model_executor
del model
gc.collect()
torch.cuda.empty_cache()
print(f"===============================")
print(f"------ END OF MultiPL-E -------")