评测结果内容为空

[yingwen330]

脚本如下： hf_vicuna_33b_v13.py
from opencompass.models import HuggingFaceCausalLM

models = [
dict(
type=HuggingFaceCausalLM,
abbr='vicuna-33b-v1.3-hf',
path="/data/models/vicuna-33b-v1.3",
tokenizer_path="/data/models/vicuna-33b-v1.3",
tokenizer_kwargs=dict(
padding_side='left',
truncation_side='left',
use_fast=False,
),
max_out_len=100,
max_seq_len=2048,
batch_size=8,
model_kwargs=dict(device_map='auto'),
batch_padding=False, # if false, inference with for-loop without batch padding
run_cfg=dict(num_gpus=4, num_procs=1)
)
]

执行脚本命令:python run.py --models hf_vicuna_33b_v13 --datasets ceval_gen

结果如下：

[SefaZeng]

同样的问题，请问解决了吗？

[mmarzl17]

解决了吗

[Leymore]

The previous log should have a hint similar to the one below. You can open and view this .out file to see the error message.

11/01 14:39:00 - OpenCompass - WARNING - task OpenICLInfer[InternLM-20B-Chat/math_19] failed, see                                                                              | 20/993 [07:03<3:41:23, 13.65s/it]
./outputs/20231031_regression/chat/20231031_225246/logs/infer/InternLM-20B-Chat/math_19.out

[SijunWang]

'torchrun' 不是内部或外部命令，也不是可运行的程序或批处理文件。这个怎么解决？

[Leymore]

install torchrun via pip3 install torch.
You may need to follow instruction from https://opencompass.readthedocs.io/en/latest/get_started/installation.html