factual_accuracy_analysis.yaml

_type: prompt
input_types: {}
input_variables:
  - context
  - input
  - format_instructions
name: null
output_parser: null
partial_variables: {}
template: |
          <|begin_of_text|><|start_header_id|>system<|end_header_id|>
          Check if the information provided by the agent during the following conversation is correct based only in the facts
          If the conversation is not related with the provided facts the score will be 100, and shouldn't be pointed any error

          <facts>
          {context}
          </facts>

          Then give an overall score of the context following of the call if there is no factual errors the score will be 100, if there are errors discount 20 points per error
          
          First do the explanation of your analysis and at the end finish with output in the requested format
          {format_instructions}
          <|eot_id|><|start_header_id|>user<|end_header_id|>

          <conversation>
          {input}
          </conversation>

          <|eot_id|><|start_header_id|>assistant<|end_header_id|>

template_format: f-string
validate_template: false