Source code for scripts.bench_utils

"""Shared utilities for benchmark and evaluation scripts."""

from __future__ import annotations

import math
import os
import statistics
from typing import Any, Optional, Sequence




def safe_mean(values: Sequence[float]) -> float:
    """Return the arithmetic mean or ``0.0`` for an empty sequence."""
    return float(statistics.fmean(values)) if values else 0.0





def percentile(values: Sequence[float], p: float) -> float:
    """Return the linear-interpolated percentile for *values*."""
    if not values:
        return 0.0
    sorted_v = sorted(values)
    k = (len(sorted_v) - 1) * (p / 100.0)
    f = math.floor(k)
    c = math.ceil(k)
    if f == c:
        return sorted_v[int(k)]
    return sorted_v[f] * (c - k) + sorted_v[c] * (k - f)





def format_float(val: float, decimals: int = 3) -> str:
    """Format a float with a fixed number of decimal places."""
    return f"{val:.{decimals}f}"





def build_ragas_embeddings(
    provider: str,
    api_key: Optional[str],
) -> Any:
    """Build a RAGAS-compatible embeddings wrapper for the judge provider."""
    try:
        from ragas.embeddings import LangchainEmbeddingsWrapper
    except ImportError:
        return None

    if provider == "mistral":
        try:
            from langchain_mistralai import MistralAIEmbeddings

            emb = MistralAIEmbeddings(
                model="mistral-embed",
                api_key=api_key or os.getenv("MISTRAL_API_KEY", ""),  # type: ignore[arg-type]
            )
            return LangchainEmbeddingsWrapper(emb)
        except ImportError:
            return None
    return None