A22635. 检索增强生成这是一个基于检索增强生成（Retrieval-Augmented Generation，RAG）的问答系统实现。该系统包含两个主要组件：1. 检索器（Retriever）：使用嵌入模型将文档转换为向量，根据查询语义检索最相关的文档2. 生成器（Generator）：基于检索到的相关文档和原始问题，生成准确的答案请根据代码逻辑和注释提示，补全代码中的空位。python import t…

检索增强生成

这是一个基于检索增强生成（Retrieval-Augmented Generation，RAG）的问答系统实现。该系统包含两个主要组件：

1. 检索器（Retriever）：使用嵌入模型将文档转换为向量，根据查询语义检索最相关的文档

2. 生成器（Generator）：基于检索到的相关文档和原始问题，生成准确的答案

请根据代码逻辑和注释提示，补全代码中的空位。

python
import torch.nn.functional as F
import torch
from transformers import AutoTokenizer, AutoModel, AutoModelForCausalLM

class Retriever:
    def __init__(self, embedder, tokenizer, corpus):
        self.embedder = embedder
        self.tokenizer = tokenizer
        self.corpus = corpus
        self.embeddings = self.build_index(corpus)
    
    def _text_to_vector(self, text_batch):
        enc = ____[21]____
        enc = {k: v.to(self.embedder.device) for k, v in enc.items(    )}
        with torch.no_grad(    ):
            outputs = ____[22]____
            last_hidden = outputs.last_hidden_state
            attn_mask = enc["attention_mask"]
            sent_vecs = ____[23]____
        return sent_vecs
    
    def build_index(self, corpus):
        vecs_tensor = self._text_to_vector(corpus)
        vecs_normalized = F.normalize(vecs_tensor, p=2, dim=1)
        return vecs_normalized.cpu(    )
    
    def search(self, query_vec, topk=3):
        if query_vec.dim(    ) == 1:
            query_vec = query_vec.unsqueeze(0)
        query_normalized = F.normalize(query_vec, p=2, dim=1)
        similarities = torch.matmul(query_normalized, self.embeddings.T)
        similarities = similarities.squeeze(0)
        topk_scores, topk_indices = torch.topk(similarities, k=min(topk, len(self.corpus)))
        results = [{"text": self.corpus[idx], "score": float(topk_scores[i])} for i, idx in enumerate(topk_indices)]
        return results

def rag_answer(query, retriever, generator, gen_tokenizer, topk=3, max_new_tokens=128):
    query_vec_tensor = retriever._text_to_vector([query])
    query_vec = query_vec_tensor.squeeze(0).cpu(    )
    docs = ____[24]____
    context = "\n".join([d["text"] for d in docs]) if docs else ""
    prompt = f"Use the context to answer.\nContext:\n{context}\n\nQ: {query}\nA:"
    inputs = gen_tokenizer(prompt, return_tensors="pt")
    inputs = {k: v.to(generator.device) for k, v in inputs.items(    )}
    pad_id = gen_tokenizer.pad_token_id if gen_tokenizer.pad_token_id is not None else gen_tokenizer.eos_token_id
    with torch.no_grad(    ):
        output_ids = generator.generate(
            **inputs,
            do_sample=True,
            temperature=0.7,
            top_p=0.9,
            max_new_tokens=max_new_tokens,
            pad_token_id=pad_id,
            eos_token_id=gen_tokenizer.eos_token_id,
        )
        new_tokens = ____[25]____
        answer = gen_tokenizer.decode(new_tokens, skip_special_tokens=True).strip(    )
    return answer

if __name__ == "__main__":
    embedder_name = "../Qwen/Qwen3-Embedding-0.6B"
    retriever_tokenizer = AutoTokenizer.from_pretrained(embedder_name)
    embedder = AutoModel.from_pretrained(embedder_name)
    generator_name = "../Qwen/Qwen3-0.6B"
    gen_tokenizer = AutoTokenizer.from_pretrained(generator_name)
    generator = AutoModelForCausalLM.from_pretrained(generator_name)
    corpus = [
        "Paris is the capital of France.",
        "Tokyo is the capital of Japan.",
        "Beijing is the capital of China.",
    ]
    retriever = Retriever(embedder, retriever_tokenizer, corpus)
    query = "What is the capital of France?"
    ans = rag_answer(query, retriever, generator, gen_tokenizer, topk=2)
    print("Q:", query)
    print("A:", ans)

在方法_text_to_vector中第一步是将输入的文本批次text_batch转换为模型可以理解的格式，以下代码正确的是：(    )