MCPcopy Create free account
hub / github.com/InternScience/SciReason / get_ppl

Method get_ppl

opencompass/models/base.py:68–86  ·  view source on GitHub ↗

Get perplexity scores given a list of inputs. Args: inputs (List[str]): A list of strings. mask_length (Optional[List[int]]): A list of mask lengths. If provided, the perplexity scores will be calculated with the first mask_length[i] t

(self,
                inputs: List[str],
                mask_length: Optional[List[int]] = None)

Source from the content-addressed store, hash-verified

66
67 @abstractmethod
68 def get_ppl(self,
69 inputs: List[str],
70 mask_length: Optional[List[int]] = None) -> List[float]:
71 """Get perplexity scores given a list of inputs.
72
73 Args:
74 inputs (List[str]): A list of strings.
75 mask_length (Optional[List[int]]): A list of mask lengths. If
76 provided, the perplexity scores will be calculated with the
77 first mask_length[i] tokens masked out. It's okay to skip
78 its implementation if advanced features in PPLInfernecer is
79 not needed.
80
81 Returns:
82 List[float]: A list of perplexity scores.
83 """
84 raise NotImplementedError(f'{self.__class__.__name__} does not support'
85 ' ppl-based evaluation yet, try gen-based '
86 'instead.')
87
88 @abstractmethod
89 def get_ppl_tokenwise(

Callers 1

get_ppl_from_templateMethod · 0.95

Calls

no outgoing calls

Tested by

no test coverage detected