Per-token log-probability evaluation result. More...

#include <entropic/types/logprob_result.h>

Public Attributes
std::vector< float >	logprobs
	Log-prob for each token transition (N-1 values)

std::vector< int32_t >	tokens
	Input tokens echoed back for verification.

float	perplexity = 0.0f
	exp(-mean(logprobs)) — lower = less surprising

float	total_logprob = 0.0f
	Sum of all logprob values.

int	n_tokens = 0
	Number of input tokens.

int	n_logprobs = 0
	Number of logprob values (n_tokens - 1)

Detailed Description

Per-token log-probability evaluation result.

Produced by InferenceBackend::evaluate_logprobs(). The base class computes perplexity and total_logprob from the raw logprobs returned by the backend implementation.

Version: 1.9.10

Definition at line 36 of file logprob_result.h.

Member Data Documentation

◆ logprobs

std::vector<float> entropic::LogprobResult::logprobs

Log-prob for each token transition (N-1 values)

Definition at line 37 of file logprob_result.h.

◆ n_logprobs

int entropic::LogprobResult::n_logprobs = 0

Number of logprob values (n_tokens - 1)

Definition at line 42 of file logprob_result.h.

◆ n_tokens

int entropic::LogprobResult::n_tokens = 0

Number of input tokens.

Definition at line 41 of file logprob_result.h.

◆ perplexity

float entropic::LogprobResult::perplexity = 0.0f

exp(-mean(logprobs)) — lower = less surprising

Definition at line 39 of file logprob_result.h.

◆ tokens

std::vector<int32_t> entropic::LogprobResult::tokens

Input tokens echoed back for verification.

Definition at line 38 of file logprob_result.h.

◆ total_logprob

float entropic::LogprobResult::total_logprob = 0.0f

Sum of all logprob values.

Definition at line 40 of file logprob_result.h.

The documentation for this struct was generated from the following file:

entropic/types/logprob_result.h

Public Attributes