mirror of
https://github.com/open-compass/opencompass.git
synced 2025-05-30 16:03:24 +08:00
add prompt postion args
This commit is contained in:
DseidLi
parent
367ba1ba61
commit
75425acdf8
@ -48,6 +48,7 @@ class NeedleBenchMultiDataset(BaseDataset):
|
|||||||
needle_file_name: str,
|
needle_file_name: str,
|
||||||
num_needles: int,
|
num_needles: int,
|
||||||
diff: int,
|
diff: int,
|
||||||
|
position: str = 'End',
|
||||||
):
|
):
|
||||||
data = {'prompt': [], 'answer': []}
|
data = {'prompt': [], 'answer': []}
|
||||||
tokenizer = tiktoken.encoding_for_model(tokenizer_model)
|
tokenizer = tiktoken.encoding_for_model(tokenizer_model)
|
||||||
@ -109,19 +110,42 @@ class NeedleBenchMultiDataset(BaseDataset):
|
|||||||
retrieval_question)
|
retrieval_question)
|
||||||
|
|
||||||
if language == 'Chinese':
|
if language == 'Chinese':
|
||||||
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
if position == 'End':
|
||||||
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
||||||
'，或重复你的回答\n'
|
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
||||||
f'用户现在给你的文档是{context}\n\n'
|
'，或重复你的回答\n'
|
||||||
f'现在请问：{retrieval_question}')
|
f'用户现在给你的文档是{context}\n\n'
|
||||||
|
f'现在请问：{retrieval_question}')
|
||||||
|
elif position == 'Start':
|
||||||
|
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
||||||
|
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
||||||
|
'，或重复你的回答\n'
|
||||||
|
f'现在请问：{retrieval_question}',
|
||||||
|
f'用户现在给你的文档是{context}\n\n')
|
||||||
|
else:
|
||||||
|
raise ValueError('Unsupported position. '
|
||||||
|
'Position must be "End" or "Start".')
|
||||||
elif language == 'English':
|
elif language == 'English':
|
||||||
prompt = ('You are an intelligent AI assistant skilled in '
|
if position == 'End':
|
||||||
'answering user questions.\n'
|
prompt = ('You are an intelligent AI assistant skilled in '
|
||||||
'Please keep your answers concise and clear. Do not'
|
'answering user questions.\n'
|
||||||
' talk about irrelevant topics or repeat your '
|
'Please keep your answers concise and clear. Do '
|
||||||
'answers.\n'
|
'not talk about irrelevant topics or repeat '
|
||||||
f'The document given to you by the user is {context}'
|
'your answers.\nThe document '
|
||||||
f'\n\nNow, the question is: {retrieval_question}')
|
f'given to you by the user is {context}\n\n'
|
||||||
|
f'Now, the question is: {retrieval_question}')
|
||||||
|
elif position == 'Start':
|
||||||
|
prompt = ('You are an intelligent AI assistant skilled in '
|
||||||
|
'answering user questions.\n'
|
||||||
|
'Please keep your answers concise and clear. Do '
|
||||||
|
'not talk about irrelevant topics or repeat '
|
||||||
|
'your answers.\n'
|
||||||
|
f'Now, the question is: {retrieval_question}'
|
||||||
|
'The document given to you by the user'
|
||||||
|
f' is {context}\n\n')
|
||||||
|
else:
|
||||||
|
raise ValueError('Unsupported position. '
|
||||||
|
'Position must be "End" or "Start".')
|
||||||
else:
|
else:
|
||||||
raise ValueError(f"Language '{language}' is not supported.")
|
raise ValueError(f"Language '{language}' is not supported.")
|
||||||
|
|
||||||
|
|||||||
@ -45,6 +45,7 @@ class NeedleBenchOriginDataset(BaseDataset):
|
|||||||
guide: bool,
|
guide: bool,
|
||||||
language: str,
|
language: str,
|
||||||
needle_file_name: str,
|
needle_file_name: str,
|
||||||
|
position: str = 'End',
|
||||||
):
|
):
|
||||||
data = {'prompt': [], 'answer': []}
|
data = {'prompt': [], 'answer': []}
|
||||||
tokenizer = tiktoken.encoding_for_model(tokenizer_model)
|
tokenizer = tiktoken.encoding_for_model(tokenizer_model)
|
||||||
@ -85,19 +86,42 @@ class NeedleBenchOriginDataset(BaseDataset):
|
|||||||
retrieval_question)
|
retrieval_question)
|
||||||
|
|
||||||
if language == 'Chinese':
|
if language == 'Chinese':
|
||||||
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
if position == 'End':
|
||||||
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
||||||
'，或重复你的回答\n'
|
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
||||||
f'用户现在给你的文档是{context}\n\n'
|
'，或重复你的回答\n'
|
||||||
f'现在请问：{retrieval_question}')
|
f'用户现在给你的文档是{context}\n\n'
|
||||||
|
f'现在请问：{retrieval_question}')
|
||||||
|
elif position == 'Start':
|
||||||
|
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
||||||
|
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
||||||
|
'，或重复你的回答\n'
|
||||||
|
f'现在请问：{retrieval_question}',
|
||||||
|
f'用户现在给你的文档是{context}\n\n')
|
||||||
|
else:
|
||||||
|
raise ValueError('Unsupported position. '
|
||||||
|
'Position must be "End" or "Start".')
|
||||||
elif language == 'English':
|
elif language == 'English':
|
||||||
prompt = ('You are an intelligent AI assistant skilled in '
|
if position == 'End':
|
||||||
'answering user questions.\n'
|
prompt = ('You are an intelligent AI assistant skilled in '
|
||||||
'Please keep your answers concise and clear. Do not'
|
'answering user questions.\n'
|
||||||
' talk about irrelevant topics or repeat your '
|
'Please keep your answers concise and clear. Do '
|
||||||
'answers.\n'
|
'not talk about irrelevant topics or repeat '
|
||||||
f'The document given to you by the user is {context}'
|
'your answers.\nThe document '
|
||||||
f'\n\nNow, the question is: {retrieval_question}')
|
f'given to you by the user is {context}\n\n'
|
||||||
|
f'Now, the question is: {retrieval_question}')
|
||||||
|
elif position == 'Start':
|
||||||
|
prompt = ('You are an intelligent AI assistant skilled in '
|
||||||
|
'answering user questions.\n'
|
||||||
|
'Please keep your answers concise and clear. Do '
|
||||||
|
'not talk about irrelevant topics or repeat '
|
||||||
|
'your answers.\n'
|
||||||
|
f'Now, the question is: {retrieval_question}'
|
||||||
|
'The document given to you by the user'
|
||||||
|
f' is {context}\n\n')
|
||||||
|
else:
|
||||||
|
raise ValueError('Unsupported position. '
|
||||||
|
'Position must be "End" or "Start".')
|
||||||
else:
|
else:
|
||||||
raise ValueError(f"Language '{language}' is not supported.")
|
raise ValueError(f"Language '{language}' is not supported.")
|
||||||
|
|
||||||
|
|||||||
@ -67,6 +67,7 @@ class NeedleBenchParallelDataset(BaseDataset):
|
|||||||
length_buffer: int,
|
length_buffer: int,
|
||||||
guide: bool,
|
guide: bool,
|
||||||
language: str,
|
language: str,
|
||||||
|
position: str = 'End',
|
||||||
):
|
):
|
||||||
data = {'prompt': [], 'answer': []}
|
data = {'prompt': [], 'answer': []}
|
||||||
tokenizer = tiktoken.encoding_for_model(tokenizer_model)
|
tokenizer = tiktoken.encoding_for_model(tokenizer_model)
|
||||||
@ -134,20 +135,41 @@ class NeedleBenchParallelDataset(BaseDataset):
|
|||||||
retrieval_question)
|
retrieval_question)
|
||||||
|
|
||||||
if language == 'Chinese':
|
if language == 'Chinese':
|
||||||
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
if position == 'End':
|
||||||
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
||||||
'，或重复你的回答\n请先仔细阅读下面的文档再依次回答'
|
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
||||||
f'最后提出的问题\n用户现在给你的文档是{context}\n\n'
|
'，或重复你的回答\n请先仔细阅读下面的文档再依次回答'
|
||||||
f'现在请问：{retrieval_question}\n')
|
f'最后提出的问题\n用户现在给你的文档是{context}\n\n'
|
||||||
|
f'现在请问：{retrieval_question}\n')
|
||||||
|
if position == 'Start':
|
||||||
|
prompt = ('你是一个善于回答用户问题的智能AI助手\n'
|
||||||
|
'请保持你的回答简洁清楚。不要说和下面文档中的无关的话'
|
||||||
|
'，或重复你的回答\n请先仔细阅读下面的文档再依次回答'
|
||||||
|
f'最后提出的问题\n现在请问：{retrieval_question}\n\n'
|
||||||
|
f'用户现在给你的文档是{context}\n')
|
||||||
|
else:
|
||||||
|
raise ValueError('Unsupported position. '
|
||||||
|
'Position must be "End" or "Start".')
|
||||||
|
|
||||||
elif language == 'English':
|
elif language == 'English':
|
||||||
prompt = (
|
if position == 'End':
|
||||||
'You are an intelligent AI assistant skilled in '
|
prompt = (
|
||||||
'answering user questions.\n'
|
'You are an intelligent AI assistant skilled in '
|
||||||
'Please keep your answers concise and clear. Do not'
|
'answering user questions.\n'
|
||||||
' talk about irrelevant topics or repeat your '
|
'Please keep your answers concise and clear. Do not'
|
||||||
'answers.\n'
|
' talk about irrelevant topics or repeat your '
|
||||||
f'The document given to you by the user is {context}'
|
'answers.\n'
|
||||||
f'\n\nNow, the questions are: {retrieval_question}\n')
|
f'The document given to you by the user is {context}'
|
||||||
|
f'\n\nNow, the questions are: {retrieval_question}\n')
|
||||||
|
if position == 'Start':
|
||||||
|
prompt = (
|
||||||
|
'You are an intelligent AI assistant skilled in '
|
||||||
|
'answering user questions.\n'
|
||||||
|
'Please keep your answers concise and clear. Do not'
|
||||||
|
' talk about irrelevant topics or repeat your '
|
||||||
|
'answers.\n'
|
||||||
|
f'\nNow, the questions are: {retrieval_question}\n\n'
|
||||||
|
f'The document given to you by the user is {context}')
|
||||||
else:
|
else:
|
||||||
raise ValueError(f"Language '{language}' is not supported.")
|
raise ValueError(f"Language '{language}' is not supported.")
|
||||||
|
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user