|
from mmengine.config import read_base |
|
from opencompass.models import OpenAI |
|
from opencompass.partitioners import NaivePartitioner |
|
from opencompass.runners import LocalRunner |
|
from opencompass.tasks import OpenICLInferTask |
|
|
|
with read_base(): |
|
from .datasets.collections.chat_medium import datasets |
|
from .summarizers.medium import summarizer |
|
|
|
|
|
from opencompass.datasets.humaneval import humaneval_gpt_postprocess |
|
for _dataset in datasets: |
|
if _dataset['path'] == 'openai_humaneval': |
|
_dataset['eval_cfg']['pred_postprocessor']['type'] = humaneval_gpt_postprocess |
|
|
|
|
|
api_meta_template = dict( |
|
round=[ |
|
dict(role='HUMAN', api_role='HUMAN'), |
|
dict(role='BOT', api_role='BOT', generate=True), |
|
], |
|
) |
|
|
|
models = [ |
|
dict(abbr='GPT4', |
|
type=OpenAI, path='gpt-4-0613', |
|
key='ENV', |
|
meta_template=api_meta_template, |
|
query_per_second=1, |
|
max_out_len=2048, max_seq_len=2048, batch_size=8), |
|
] |
|
|
|
infer = dict( |
|
partitioner=dict(type=NaivePartitioner), |
|
runner=dict( |
|
type=LocalRunner, |
|
max_num_workers=4, |
|
task=dict(type=OpenICLInferTask)), |
|
) |
|
|