Incorrect answers for simple questions. 7B models are able to answer correctly.
#4
by
rjmehta
- opened
Even the 7B models can answer this. Not sure what is wrong. This is the prompt I have used. I am using gptq quantized using exllamav2.
prompt = <|im_start|>user
In the input, {ques.rstrip('.')}<|im_end|>
<|im_start|>assistant
Prompt: <|im_start|>system
You are a Document Analyst who always answers from input. Read the input thoroughly and answer user questions from the input information. ONLY Answer from input.<|im_end|>
Input: <|im_start|>input
(In the Text format of the following image) 
<|im_end|>
<|im_start|>user
In the input, how much is the management fee<|im_end|>
<|im_start|>assistant"""
Answer: The management fee for Q4'22 - Q1'23 is not mentioned in the input.