我要投稿

RAG 数据集准备篇：用三大 Agents 优化评估流程

发布日期：2024-12-15 18:03:29 浏览次数： 2282 作者：非架构

本文是关于如何优化 RAG 技术的一系列文章之一。在之前的文章中，我们已经深入探讨了如何在 Chunking、Embedding 以及评估指标设计等环节优化 RAG 性能。这篇文章将重点关注如何准备评估 RAG 性能所需的数据集，为后续优化打下坚实基础。

1. 优化 RAG 不是“炼丹”，需要系统的方法

当涉及到如何优化 RAG（Retrieval-Augmented Generation）性能时，许多人往往采取“试错”的方法：“我尝试了一个新模块，问了几个问题，答案看起来还行…”。这样的定性评估方式虽然直观，但很难得出可靠的改进结论。

优化 RAG 性能需要采用科学实验的方式。这意味着我们必须设计量化的评估指标，并准备高质量的评估数据集。通过结构化的实验和评估方法，我们才能明确什么样的调整会真正提升性能。

随机提问题这种太上老君“炼丹”的方法不适用于 RAG 性能评估和优化

2. 结构化评估 RAG 的性能

RAG 系统的性能评估可以从以下两个主要方面进行：

检索评估（Retrieval Evaluation）： 检索的段落是否相关？
生成评估（Generation Evaluation）： 模型生成的答案是否恰当？

对于端到端的评估，我们通常关注以下四类指标：

Groundedness（可靠性）： 检索评估中的关键指标，评估检索的段落是否为生成的答案提供了可靠支持，避免幻觉（hallucinations）或无关信息。这是生成评估的基础。
Completeness（完整性）： 生成评估的核心指标，用于评估模型的回答是否全面覆盖了用户问题的所有方面，同时间接反映了检索阶段的信息充分性。
Utilization（利用率）： 连接检索和生成的桥梁指标，用来评估检索到的信息是否被有效利用。如果利用率低，可能意味着检索到的段落与问题无关，或者模型未充分使用这些段落。
Relevance（相关性）： 检索评估的直接反映，衡量检索到的段落与用户问题之间的相关性，同时也会影响生成评估的整体表现。

这些指标需要基于高质量的测试样本和严格的评估流程。

3. 使用 Agents 完成 RAG 性能基准测试

在开始具体流程前，我们需要了解三个代理（Agents）的协作关系：

测试样本代理（Test Sample Agent）： 用于生成和准备测试数据，这些数据构成了评估系统的基础。
样本质量评价代理（Critique Agent）： 对测试样本进行质量审核，确保最终用于评估的数据具备高准确性和清晰性。
评估代理（Evaluation Agent）： 负责根据设计的指标对系统性能进行量化评估。这些评估指标覆盖了检索和生成的多个层面。

通过这三种代理的协作，可以建立一个系统化的性能基准测试流程，帮助开发者深入理解和优化 RAG 系统。

3.1 生成测试样本

为了评估 RAG 系统的性能，首先需要使用测试样本代理（Test Sample Agent） 生成测试样本。这些代理可以自动生成一组高质量的 QA 样本（问题与答案对）。例如，可以先生成 10 个样本用于快速测试，然后从公共资源库加载更多样本，或者生成更大规模的数据集以供全面评估。

但如果针对特定的知识库，建议生成至少 200 个样本，因为后续通过评价代理过滤掉低质量问题后，最终能保留的有效样本约为一半。

以下是生成QA样本的提示词：

QA_generation_prompt = """
Your task is to write a factoid question and an answer given a context.
Your factoid question should be answerable with a specific, concise piece of factual information from the context.
Your factoid question should be formulated in the same style as questions users could ask in a search engine.
This means that your factoid question MUST NOT mention something like "according to the passage" or "context".

Provide your answer as follows:

Output:::
Factoid question: (your factoid question)
Answer: (your answer to the factoid question)

Now here is the context.

Context: {context}\n
Output:::"""

3.2 检查样本质量

自动生成的样本可能存在质量问题，因此需要引入样本质量评价代理（Critique Agents） 来对问题进行质量审查。这些代理会根据多种标准对每个问题评分，比如：

问题是否清晰、无二义性？
问题是否适合特定知识领域？

我们通过这些代理对问题进行系统评分。当任意一个代理的评分过低时，直接剔除该问题。

?提示： 当让代理生成分数时，先要求其输出理由，再给出最终评分。这种方式不仅能帮助我们验证评分结果，还能促使代理在回答过程中进行更深入的思考，从而提高评分的准确性。

以下是评估样本质量的提示词：

question_groundedness_critique_prompt = """
You will be given a context and a question.
Your task is to provide a 'total rating' scoring how well one can answer the given question unambiguously with the given context.
Give your answer on a scale of 1 to 5, where 1 means that the question is not answerable at all given the context, and 5 means that the question is clearly and unambiguously answerable with the context.

Provide your answer as follows:

Answer:::
Evaluation: (your rationale for the rating, as a text)
Total rating: (your rating, as a number between 1 and 5)

You MUST provide values for 'Evaluation:' and 'Total rating:' in your answer.

Now here are the question and context.

Question: {question}\n
Context: {context}\n
Answer::: """

question_relevance_critique_prompt = """
You will be given a question.
Your task is to provide a 'total rating' representing how useful this question can be to machine learning developers building NLP applications with the Hugging Face ecosystem.
Give your answer on a scale of 1 to 5, where 1 means that the question is not useful at all, and 5 means that the question is extremely useful.

Provide your answer as follows:

Answer:::
Evaluation: (your rationale for the rating, as a text)
Total rating: (your rating, as a number between 1 and 5)

You MUST provide values for 'Evaluation:' and 'Total rating:' in your answer.

Now here is the question.

Question: {question}\n
Answer::: """

question_standalone_critique_prompt = """
You will be given a question.
Your task is to provide a 'total rating' representing how context-independant this question is.
Give your answer on a scale of 1 to 5, where 1 means that the question depends on additional information to be understood, and 5 means that the question makes sense by itself.
For instance, if the question refers to a particular setting, like 'in the context' or 'in the document', the rating must be 1.
The questions can contain obscure technical nouns or acronyms like Gradio, Hub, Hugging Face or Space and still be a 5: it must simply be clear to an operator with access to documentation what the question is about.

For instance, "What is the name of the checkpoint from which the ViT model is imported?" should receive a 1, since there is an implicit mention of a context, thus the question is not independant from the context.

Provide your answer as follows:

Answer:::
Evaluation: (your rationale for the rating, as a text)
Total rating: (your rating, as a number between 1 and 5)

You MUST provide values for 'Evaluation:' and 'Total rating:' in your answer.

Now here is the question.

Question: {question}\n
Answer::: """

3.3 设置评估代理

最后，我们需要评估 RAG 系统在测试数据集上的表现。可以通过以下步骤完成：

选择评估指标：

我们重点关注faithfulness（可靠性） 作为主要指标，因为它能全面反映系统的端到端性能。

选择评估模型：

使用 GPT-4 作为评估代理，或者尝试其他性能良好的模型，如 kaist-ai/prometheus-13b-v1.0 或 BAAI/JudgeLM-33B-v1.0。

设计评估提示词：

提示词需要详细描述每个指标的评分标准（例如 1-5 分），并要求模型在评分前先输出评分依据。

?提示： 提供详细的评分标准有助于评估代理保持一致性，避免因模糊标准导致评分结果的波动。

以下是 RAG 评估模型的提示词：

EVALUATION_PROMPT = """###Task Description:
An instruction (might include an Input inside it), a response to evaluate, a reference answer that gets a score of 5, and a score rubric representing a evaluation criteria are given.
1. Write a detailed feedback that assess the quality of the response strictly based on the given score rubric, not evaluating in general.
2. After writing a feedback, write a score that is an integer between 1 and 5. You should refer to the score rubric.
3. The output format should look as follows: \"Feedback: {{write a feedback for criteria}} [RESULT] {{an integer number between 1 and 5}}\"
4. Please do not generate any other opening, closing, and explanations. Be sure to include [RESULT] in your output.

###The instruction to evaluate:
{instruction}

###Response to evaluate:
{response}

###Reference Answer (Score 5):
{reference_answer}

###Score Rubrics:
[Is the response correct, accurate, and factual based on the reference answer?]
Score 1: The response is completely incorrect, inaccurate, and/or not factual.
Score 2: The response is mostly incorrect, inaccurate, and/or not factual.
Score 3: The response is somewhat correct, accurate, and/or factual.
Score 4: The response is mostly correct, accurate, and factual.
Score 5: The response is completely correct, accurate, and factual.

###Feedback:"""

from langchain.prompts.chat import (
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
)
from langchain.schema import SystemMessage


evaluation_prompt_template = ChatPromptTemplate.from_messages(
    [
        SystemMessage(content="You are a fair evaluator language model."),
        HumanMessagePromptTemplate.from_template(EVALUATION_PROMPT),
    ]
)