OpenCompass/opencompass/configs/datasets/longbench/longbenchtriviaqa/longbench_triviaqa_gen_d30cb9.py
2024-09-06 15:50:12 +08:00

50 lines
1.5 KiB
Python

from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.datasets import (
LongBenchF1Evaluator,
LongBenchtriviaqaDataset,
triviaqa_postprocess,
)
LongBench_triviaqa_reader_cfg = dict(
input_columns=['context', 'input'],
output_column='answers',
train_split='test',
test_split='test',
)
LongBench_triviaqa_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(
role='HUMAN',
prompt='Answer the question based on the given passage. Only give me the answer and do not output any other words. The following are some examples.\n\n{context}\n\n{input}',
),
],
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=32),
)
LongBench_triviaqa_eval_cfg = dict(
evaluator=dict(type=LongBenchF1Evaluator),
pred_role='BOT',
pred_postprocessor=dict(type=triviaqa_postprocess),
)
LongBench_triviaqa_datasets = [
dict(
type=LongBenchtriviaqaDataset,
abbr='LongBench_triviaqa',
path='opencompass/Longbench',
name='triviaqa',
reader_cfg=LongBench_triviaqa_reader_cfg,
infer_cfg=LongBench_triviaqa_infer_cfg,
eval_cfg=LongBench_triviaqa_eval_cfg,
)
]