kscz0000 LLM Evaluation
掌握 LLM 应用的全面评估策略，涵盖自动化指标、人工评估和 A/B 测试。
何时不使用此技能
- 任务与 LLM 评估无关
- 需要此范围之外的其他领域或工具
使用说明
- 明确目标、约束和所需输入。
- 应用相关最佳实践并验证结果。
- 提供可操作的步骤和验证方法。
- 如需详细示例，打开
resources/implementation-playbook.md。
何时使用此技能
- 系统化衡量 LLM 应用性能
- 比较不同模型或提示词
- 部署前检测性能回归
- 验证提示词变更带来的改进
- 建立对生产系统的信心
- 建立基线并随时间追踪进展
- 调试异常的模型行为
核心评估类型
1. 自动化指标
快速、可重复、可扩展的评估，基于计算得分。
文本生成：
- BLEU：N-gram 重叠度（翻译）
- ROUGE：面向召回率（摘要）
- METEOR：语义相似度
- BERTScore：基于嵌入的相似度
- Perplexity：语言模型置信度
分类：
- Accuracy：正确率
- Precision/Recall/F1：按类别衡量的性能
- Confusion Matrix：错误模式
- AUC-ROC：排序质量
检索（RAG）：
- MRR：平均倒数排名
- NDCG：归一化折损累积增益
- Precision@K：前 K 个中的相关结果
- Recall@K：前 K 个中的覆盖度
2. 人工评估
针对难以自动化的质量维度进行人工评定。
维度：
- Accuracy：事实正确性
- Coherence：逻辑连贯性
- Relevance：是否回答了问题
- Fluency：语言自然度
- Safety：无有害内容
- Helpfulness：对用户有用
3. LLM-as-Judge
用更强的 LLM 评估较弱模型的输出。
方法：
- Pointwise：对单个响应打分
- Pairwise：比较两个响应
- Reference-based：与黄金标准对比
- Reference-free：无真实标签评判
快速开始
from llm_eval import EvaluationSuite, Metric
# Define evaluation suite
suite = EvaluationSuite([
Metric.accuracy(),
Metric.bleu(),
Metric.bertscore(),
Metric.custom(name="groundedness", fn=check_groundedness)
])
# Prepare test cases
test_cases = [
{
"input": "What is the capital of France?",
"expected": "Paris",
"context": "France is a country in Europe. Paris is its capital."
},
# ... more test cases
]
# Run evaluation
results = suite.evaluate(
model=your_model,
test_cases=test_cases
)
print(f"Overall Accuracy: {results.metrics['accuracy']}")
print(f"BLEU Score: {results.metrics['bleu']}")
自动化指标实现
BLEU 分数
from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
def calculate_bleu(reference, hypothesis):
"""Calculate BLEU score between reference and hypothesis."""
smoothie = SmoothingFunction().method4
return sentence_bleu(
[reference.split()],
hypothesis.split(),
smoothing_function=smoothie
)
# Usage
bleu = calculate_bleu(
reference="The cat sat on the mat",
hypothesis="A cat is sitting on the mat"
)
ROUGE 分数
from rouge_score import rouge_scorer
def calculate_rouge(reference, hypothesis):
"""Calculate ROUGE scores."""
scorer = rouge_scorer.RougeScorer(['rouge1', 'rouge2', 'rougeL'], use_stemmer=True)
scores = scorer.score(reference, hypothesis)
return {
'rouge1': scores['rouge1'].fmeasure,
'rouge2': scores['rouge2'].fmeasure,
'rougeL': scores['rougeL'].fmeasure
}
BERTScore
from bert_score import score
def calculate_bertscore(references, hypotheses):
"""Calculate BERTScore using pre-trained BERT."""
P, R, F1 = score(
hypotheses,
references,
lang='en',
model_type='microsoft/deberta-xlarge-mnli'
)
return {
'precision': P.mean().item(),
'recall': R.mean().item(),
'f1': F1.mean().item()
}
自定义指标
def calculate_groundedness(response, context):
"""Check if response is grounded in provided context."""
# Use NLI model to check entailment
from transformers import pipeline
nli = pipeline("text-classification", model="microsoft/deberta-large-mnli")
result = nli(f"{context} [SEP] {response}")[0]
# Return confidence that response is entailed by context
return result['score'] if result['label'] == 'ENTAILMENT' else 0.0
def calculate_toxicity(text):
"""Measure toxicity in generated text."""
from detoxify import Detoxify
results = Detoxify('original').predict(text)
return max(results.values()) # Return highest toxicity score
def calculate_factuality(claim, knowledge_base):
"""Verify factual claims against knowledge base."""
# Implementation depends on your knowledge base
# Could use retrieval + NLI, or fact-checking API
pass
LLM-as-Judge 模式
单输出评估
def llm_judge_quality(response, question):
"""Use GPT-5 to judge response quality."""
prompt = f"""Rate the following response on a scale of 1-10 for:
1. Accuracy (factually correct)
2. Helpfulness (answers the question)
3. Clarity (well-written and understandable)
Question: {question}
Response: {response}
Provide ratings in JSON format:
{{
"accuracy": <1-10>,
"helpfulness": <1-10>,
"clarity": <1-10>,
"reasoning": "<brief explanation>"
}}
"""
result = openai.ChatCompletion.create(
model="gpt-5",
messages=[{"role": "user", "content": prompt}],
temperature=0
)
return json.loads(result.choices[0].message.content)
成对比较
def compare_responses(question, response_a, response_b):
"""Compare two responses using LLM judge."""
prompt = f"""Compare these two responses to the question and determine which is better.
Question: {question}
Response A: {response_a}
Response B: {response_b}
Which response is better and why? Consider accuracy, helpfulness, and clarity.
Answer with JSON:
{{
"winner": "A" or "B" or "tie",
"reasoning": "<explanation>",
"confidence": <1-10>
}}
"""
result = openai.ChatCompletion.create(
model="gpt-5",
messages=[{"role": "user", "content": prompt}],
temperature=0
)
return json.loads(result.choices[0].message.content)
人工评估框架
标注指南
class AnnotationTask:
"""Structure for human annotation task."""
def __init__(self, response, question, context=None):
self.response = response
self.question = question
self.context = context
def get_annotation_form(self):
return {
"question": self.question,
"context": self.context,
"response": self.response,
"ratings": {
"accuracy": {
"scale": "1-5",
"description": "Is the response factually correct?"
},
"relevance": {
"scale": "1-5",
"description": "Does it answer the question?"
},
"coherence": {
"scale": "1-5",
"description": "Is it logically consistent?"
}
},
"issues": {
"factual_error": False,
"hallucination": False,
"off_topic": False,
"unsafe_content": False
},
"feedback": ""
}
评分者间一致性
from sklearn.metrics import cohen_kappa_score
def calculate_agreement(rater1_scores, rater2_scores):
"""Calculate inter-rater agreement."""
kappa = cohen_kappa_score(rater1_scores, rater2_scores)
interpretation = {
kappa < 0: "Poor",
kappa < 0.2: "Slight",
kappa < 0.4: "Fair",
kappa < 0.6: "Moderate",
kappa < 0.8: "Substantial",
kappa <= 1.0: "Almost Perfect"
}
return {
"kappa": kappa,
"interpretation": interpretation[True]
}
A/B 测试
统计检验框架
from scipy import stats
import numpy as np
class ABTest:
def __init__(self, variant_a_name="A", variant_b_name="B"):
self.variant_a = {"name": variant_a_name, "scores": []}
self.variant_b = {"name": variant_b_name, "scores": []}
def add_result(self, variant, score):
"""Add evaluation result for a variant."""
if variant == "A":
self.variant_a["scores"].append(score)
else:
self.variant_b["scores"].append(score)
def analyze(self, alpha=0.05):
"""Perform statistical analysis."""
a_scores = self.variant_a["scores"]
b_scores = self.variant_b["scores"]
# T-test
t_stat, p_value = stats.ttest_ind(a_scores, b_scores)
# Effect size (Cohen's d)
pooled_std = np.sqrt((np.std(a_scores)**2 + np.std(b_scores)**2) / 2)
cohens_d = (np.mean(b_scores) - np.mean(a_scores)) / pooled_std
return {
"variant_a_mean": np.mean(a_scores),
"variant_b_mean": np.mean(b_scores),
"difference": np.mean(b_scores) - np.mean(a_scores),
"relative_improvement": (np.mean(b_scores) - np.mean(a_scores)) / np.mean(a_scores),
"p_value": p_value,
"statistically_significant": p_value < alpha,
"cohens_d": cohens_d,
"effect_size": self.interpret_cohens_d(cohens_d),
"winner": "B" if np.mean(b_scores) > np.mean(a_scores) else "A"
}
@staticmethod
def interpret_cohens_d(d):
"""Interpret Cohen's d effect size."""
abs_d = abs(d)
if abs_d < 0.2:
return "negligible"
elif abs_d < 0.5:
return "small"
elif abs_d < 0.8:
return "medium"
else:
return "large"
回归测试
回归检测
class RegressionDetector:
def __init__(self, baseline_results, threshold=0.05):
self.baseline = baseline_results
self.threshold = threshold
def check_for_regression(self, new_results):
"""Detect if new results show regression."""
regressions = []
for metric in self.baseline.keys():
baseline_score = self.baseline[metric]
new_score = new_results.get(metric)
if new_score is None:
continue
# Calculate relative change
relative_change = (new_score - baseline_score) / baseline_score
# Flag if significant decrease
if relative_change < -self.threshold:
regressions.append({
"metric": metric,
"baseline": baseline_score,
"current": new_score,
"change": relative_change
})
return {
"has_regression": len(regressions) > 0,
"regressions": regressions
}
基准测试
运行基准测试
class BenchmarkRunner:
def __init__(self, benchmark_dataset):
self.dataset = benchmark_dataset
def run_benchmark(self, model, metrics):
"""Run model on benchmark and calculate metrics."""
results = {metric.name: [] for metric in metrics}
for example in self.dataset:
# Generate prediction
prediction = model.predict(example["input"])
# Calculate each metric
for metric in metrics:
score = metric.calculate(
prediction=prediction,
reference=example["reference"],
context=example.get("context")
)
results[metric.name].append(score)
# Aggregate results
return {
metric: {
"mean": np.mean(scores),
"std": np.std(scores),
"min": min(scores),
"max": max(scores)
}
for metric, scores in results.items()
}
资源
- references/metrics.md：综合指标指南
- references/human-evaluation.md：标注最佳实践
- references/benchmarking.md：标准基准测试
- references/a-b-testing.md：统计检验指南
- references/regression-testing.md：CI/CD 集成
- assets/evaluation-framework.py：完整评估框架
- assets/benchmark-dataset.jsonl：示例数据集
- scripts/evaluate-model.py：自动化评估运行器
最佳实践
- 多指标并用：使用多种指标获得全面视角
- 代表性数据：在真实、多样的样本上测试
- 基线对比：始终与基线性能比较
- 统计严谨性：比较时使用适当的统计检验
- 持续评估：集成到 CI/CD 流水线
- 人工验证：将自动化指标与人工判断结合
- 错误分析：调查失败案例以理解薄弱环节
- 版本控制：随时间追踪评估结果
常见陷阱
- 单一指标执念：为优化一个指标而牺牲其他指标
- 样本量过小：从过少的样本中得出结论
- 数据污染：在训练数据上测试
- 忽略方差：未考虑统计不确定性
- 指标错配：使用的指标与业务目标不一致
局限性
- 仅在任务明确符合上述范围时使用此技能。
- 输出不能替代针对具体环境的验证、测试或专家评审。
- 如缺少必要输入、权限、安全边界或成功标准，请停下来询问澄清。