Haystack 2.x Retrievers API 深度解析從 BM25、向量檢索到句子窗口與自動(dòng)合并【免費(fèi)下載鏈接】haystackOpen-source AI orchestration framework for building context-engineered, production-ready LLM applications. Design modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation. Built for scalable agents, RAG, multimodal applications, semantic search, and conversational systems.項(xiàng)目地址: https://gitcode.com/GitHub_Trending/ha/haystack本篇技術(shù)指南以 Haystack 2.x 的 Retrievers API 文檔docs-website/reference_versioned_docs/version-2.18/haystack-api/retrievers_api.md為核心骨架系統(tǒng)講解AutoMergingRetriever、InMemoryBM25Retriever、InMemoryEmbeddingRetriever、FilterRetriever與SentenceWindowRetriever五大組件的設(shè)計(jì)原理、完整參數(shù)語(yǔ)義、代碼級(jí)調(diào)用鏈與實(shí)戰(zhàn)用法。讀完本文你將能夠根據(jù)檢索場(chǎng)景關(guān)鍵詞、語(yǔ)義、過(guò)濾、父子文檔合并、上下文窗口擴(kuò)展正確選型并組裝可運(yùn)行的 Haystack 檢索流水線。檢索器在 Haystack 中的定位在 Haystack 中Retriever檢索器是一個(gè)掃過(guò) Document Store、返回與查詢(xún)相關(guān)的候選文檔集合的組件。它與Reader等組件的核心區(qū)別在于檢索器只負(fù)責(zé)召回候選不做精讀與答案生成它通常位于 RAG 流水線的前端將召回結(jié)果交給后續(xù)的 Prompt Builder、Generator 或 Evaluator 消費(fèi)。從當(dāng)前倉(cāng)庫(kù)源碼看所有內(nèi)置檢索器都位于 haystack/components/retrievers/并通過(guò)component裝飾器注冊(cè)為可序列化、可連線的管道組件auto_merging_retriever.py —AutoMergingRetrieverin_memory/bm25_retriever.py —InMemoryBM25Retrieverin_memory/embedding_retriever.py —InMemoryEmbeddingRetrieverfilter_retriever.py —FilterRetrieversentence_window_retriever.py —SentenceWindowRetriever此外還有MultiQueryEmbeddingRetriever、MultiQueryTextRetriever、MultiRetriever、TextEmbeddingRetriever等組件見(jiàn) haystack/components/retrievers/init.py本文聚焦 API 文檔中展開(kāi)講解的五個(gè)組件。InMemoryBM25Retriever基于關(guān)鍵詞的經(jīng)典檢索InMemoryBM25Retriever使用 BM25 關(guān)鍵詞算法檢索與查詢(xún)最相似的文檔必須搭配InMemoryDocumentStore使用。最小可用示例API 文檔給出的最小示例完整可運(yùn)行from haystack import Document from haystack.components.retrievers.in_memory import InMemoryBM25Retriever from haystack.document_stores.in_memory import InMemoryDocumentStore docs [ Document(contentPython is a popular programming language), Document(contentpython ist eine beliebte Programmiersprache), ] doc_store InMemoryDocumentStore() doc_store.write_documents(docs) retriever InMemoryBM25Retriever(doc_store) result retriever.run(queryProgrammiersprache) print(result[documents])構(gòu)造參數(shù)與底層校驗(yàn)構(gòu)造函數(shù)簽名見(jiàn) in_memory/bm25_retriever.pydef __init__(document_store: InMemoryDocumentStore, filters: Optional[dict[str, Any]] None, top_k: int 10, scale_score: bool False, filter_policy: FilterPolicy FilterPolicy.REPLACE)參數(shù)默認(rèn)值語(yǔ)義document_store必填搜索目標(biāo)必須是InMemoryDocumentStore實(shí)例否則拋TypeError源碼第 70-71 行filtersNone用于縮小搜索空間的過(guò)濾字典元數(shù)據(jù)過(guò)濾top_k10返回的最大文檔數(shù)必須 0否則拋ValueError源碼第 75-76 行scale_scoreFalse為T(mén)rue時(shí)將得分歸一化到 0-11 表示極相關(guān)為False返回原始相似度得分filter_policyREPLACE過(guò)濾策略見(jiàn)下文FilterPolicy 詳解run 與 run_async運(yùn)行期覆蓋初始化參數(shù)run方法簽名component.output_types(documentslist[Document]) def run(query: str, filters: Optional[dict[str, Any]] None, top_k: Optional[int] None, scale_score: Optional[bool] None)run_async擁有完全一致的簽名用于異步執(zhí)行。底層調(diào)用鏈非常直接源碼第 146-153 行先通過(guò)apply_filter_policy(self.filter_policy, self.filters, filters)依據(jù)策略合并/替換初始化過(guò)濾器與運(yùn)行期過(guò)濾器top_k、scale_score若未在run中傳入則回退到構(gòu)造函數(shù)中的值最終委托給document_store.bm25_retrieval(query..., filters..., top_k..., scale_score...)。bm25_retrieval是InMemoryDocumentStore的原生方法haystack/document_stores/in_memory/document_store.py異步版本為bm25_retrieval_async同文件第 1054 行附近由run_async調(diào)用。也就是說(shuō)同步/異步只發(fā)生在 Document Store 層檢索器本身是薄封裝。InMemoryEmbeddingRetriever語(yǔ)義向量檢索InMemoryEmbeddingRetriever檢索與查詢(xún)?cè)谡Z(yǔ)義上最相似的文檔同樣只搭配InMemoryDocumentStore。使用前必須保證文檔與查詢(xún)兩側(cè)都有 embedding 可用索引流水線中用DocumentEmbedder為文檔生成向量查詢(xún)流水線中用TextEmbedder為查詢(xún)生成向量再傳給檢索器。完整用法示例Sentence Transformers 版API 文檔給出的完整示例使用SentenceTransformersDocumentEmbedder/SentenceTransformersTextEmbedderfrom haystack import Document from haystack.components.embedders import SentenceTransformersDocumentEmbedder, SentenceTransformersTextEmbedder from haystack.components.retrievers.in_memory import InMemoryEmbeddingRetriever from haystack.document_stores.in_memory import InMemoryDocumentStore docs [ Document(contentPython is a popular programming language), Document(contentpython ist eine beliebte Programmiersprache), ] doc_embedder SentenceTransformersDocumentEmbedder() doc_embedder.warm_up() docs_with_embeddings doc_embedder.run(docs)[documents] doc_store InMemoryDocumentStore() doc_store.write_documents(docs_with_embeddings) retriever InMemoryEmbeddingRetriever(doc_store) queryProgrammiersprache text_embedder SentenceTransformersTextEmbedder() text_embedder.warm_up() query_embedding text_embedder.run(query)[embedding] result retriever.run(query_embeddingquery_embedding) print(result[documents])注意與 BM25 的關(guān)鍵差異run的入?yún)⑹莙uery_embedding向量列表而非query字符串向量化完全由外部 Embedder 負(fù)責(zé)。構(gòu)造參數(shù)與 run 參數(shù)構(gòu)造函數(shù)in_memory/embedding_retriever.pydef __init__(document_store: InMemoryDocumentStore, filters: Optional[dict[str, Any]] None, top_k: int 10, scale_score: bool False, return_embedding: bool False, filter_policy: FilterPolicy FilterPolicy.REPLACE)參數(shù)默認(rèn)值語(yǔ)義return_embeddingFalse為T(mén)rue時(shí)檢索結(jié)果文檔中附帶各自的 embedding為False時(shí)僅返回文檔本體減少內(nèi)存占用與傳輸開(kāi)銷(xiāo)run簽名component.output_types(documentslist[Document]) def run(query_embedding: list[float], filters: Optional[dict[str, Any]] None, top_k: Optional[int] None, scale_score: Optional[bool] None, return_embedding: Optional[bool] None)run_async簽名與之完全相同。底層委托鏈源碼第 166-182 行為apply_filter_policy(...)→ 參數(shù)回退 →document_store.embedding_retrieval(query_embedding..., filters..., top_k..., scale_score..., return_embedding...)異步版本調(diào)用embedding_retrieval_asyncdocument_store.py。構(gòu)造函數(shù)與 BM25 版一致地對(duì)document_store做類(lèi)型檢查、對(duì)top_k 0拋ValueError。FilterPolicy 詳解REPLACE 與 MERGEFilterPolicy是 Haystack 檢索器統(tǒng)一的過(guò)濾策略枚舉定義于 haystack/document_stores/types/filter_policy.pyREPLACE默認(rèn)運(yùn)行期傳入的filters直接覆蓋初始化時(shí)的過(guò)濾器。適用于需要針對(duì)不同查詢(xún)動(dòng)態(tài)切換過(guò)濾條件的場(chǎng)景。MERGE運(yùn)行期過(guò)濾器與初始化過(guò)濾器合并重疊字段以運(yùn)行期值為準(zhǔn)從而進(jìn)一步收窄搜索范圍。合并邏輯由apply_filter_policy(filter_policy, init_filters, runtime_filters, default_logical_operator)實(shí)現(xiàn)同文件第 287 行起在MERGE模式下若兩側(cè)都是比較型過(guò)濾器則用combine_two_comparison_filters組合為邏輯表達(dá)式否則按默認(rèn)邏輯操作符合并。由于filter_policy在to_dict中被序列化為字符串值、在from_dict中通過(guò)FilterPolicy.from_str還原見(jiàn)兩個(gè) InMemory 檢索器的from_dict因此該配置可以完整地在 YAML/JSON 管道描述中往返。FilterRetriever純?cè)獢?shù)據(jù)過(guò)濾檢索FilterRetriever不計(jì)算任何相似度只按過(guò)濾器條件從 Document Store 中取出匹配的文檔。它接受通用的DocumentStore不僅限 InMemory是搭建先過(guò)濾、再檢索流水線的輕量工具。用法示例與運(yùn)行期覆蓋API 文檔示例from haystack import Document from haystack.components.retrievers import FilterRetriever from haystack.document_stores.in_memory import InMemoryDocumentStore docs [ Document(contentPython is a popular programming language, meta{lang: en}), Document(contentpython ist eine beliebte Programmiersprache, meta{lang: de}), ] doc_store InMemoryDocumentStore() doc_store.write_documents(docs) retriever FilterRetriever(doc_store, filters{field: lang, operator: , value: en}) # 若在 run 中傳入 filters將覆蓋初始化時(shí)的過(guò)濾器 result retriever.run(filters{field: lang, operator: , value: de}) print(result[documents])__init__(document_store, filtersNone)與run(filtersNone)的參數(shù)語(yǔ)義完全一致從源碼filter_retriever.py看run中的解析規(guī)則是resolved_filters filters if filters is not None else self.filters——運(yùn)行期過(guò)濾器一旦給出就整體覆蓋初始化值與FilterPolicy.REPLACE行為一致但該組件自身不攜帶 FilterPolicy 枚舉。異步版run_async委托filter_documents_async同步版委托filter_documents。過(guò)濾器語(yǔ)法Comparison 與 Logic過(guò)濾器是嵌套字典支持兩種類(lèi)型規(guī)范見(jiàn) haystack/document_stores/types/protocol.py比較型Comparison必須含field、operator、value三個(gè)鍵operator取值、!、、、、、in、not in。邏輯型Logic必須含operator與conditionsoperator取值NOT、OR、ANDconditions為比較型或邏輯型字典的列表。簡(jiǎn)單示例{field: meta.type, operator: , value: article}。復(fù)合示例filters { operator: AND, conditions: [ {field: meta.type, operator: , value: article}, {field: meta.date, operator: , value: 1420066800}, {field: meta.date, operator: , value: 1609455600}, {field: meta.rating, operator: , value: 3}, { operator: OR, conditions: [ {field: meta.genre, operator: in, value: [economy, politics]}, {field: meta.publisher, operator: , value: nytimes}, ], }, ], }SentenceWindowRetriever句子窗口上下文擴(kuò)展SentenceWindowRetriever從 Document Store 中取回檢索命中文檔的鄰近分塊為查詢(xún)結(jié)果補(bǔ)充上下文。它設(shè)計(jì)為接在某個(gè)基礎(chǔ) Retriever如 BM25、Embedding Retriever之后使用輸出context_windows合并后的上下文文本列表與context_documents含鄰近塊的文檔列表兩類(lèi)結(jié)果。前置條件分塊元數(shù)據(jù)約定使用該組件的前提是文檔攜帶兩塊元數(shù)據(jù)source_id標(biāo)識(shí)分塊所屬的原始文檔用于將同一來(lái)源的句子分塊歸組split_id分塊在原始文檔中的位置/順序。這兩個(gè)字段名可分別通過(guò)source_id_meta_field與split_id_meta_field自定義。與 DocumentSplitter 組合的端到端示例API 文檔給出完整流水線示例DocumentSplitter生成的source_id/split_id/split_idx_start元數(shù)據(jù)正好滿(mǎn)足約定from haystack import Document, Pipeline from haystack.components.retrievers.in_memory import InMemoryBM25Retriever from haystack.components.retrievers import SentenceWindowRetriever from haystack.components.preprocessors import DocumentSplitter from haystack.document_stores.in_memory import InMemoryDocumentStore splitter DocumentSplitter(split_length10, split_overlap5, split_byword) text ( This is a text with some words. There is a second sentence. And there is also a third sentence. It also contains a fourth sentence. And a fifth sentence. And a sixth sentence. And a seventh sentence ) doc Document(contenttext) docs splitter.run([doc]) doc_store InMemoryDocumentStore() doc_store.write_documents(docs[documents]) rag Pipeline() rag.add_component(bm25_retriever, InMemoryBM25Retriever(doc_store, top_k1)) rag.add_component(sentence_window_retriever, SentenceWindowRetriever(document_storedoc_store, window_size2)) rag.connect(bm25_retriever, sentence_window_retriever) rag.run({bm25_retriever: {query:third}})運(yùn)行后sentence_window_retriever的輸出形如{sentence_window_retriever: {context_windows: [some words. There is a second sentence. And there is also a third sentence. It also contains a fourth sentence. And a fifth sentence. And a sixth sentence. And a], context_documents: [Document(id..., content: some words. There is a second sentence. And there is , meta: {source_id: ..., page_number: 1, split_id: 1, split_idx_start: 20, ...}), Document(id..., content: second sentence. And there is also a third sentence. It , meta: {source_id: ..., split_id: 2, split_idx_start: 43, ...}), ...]}}命中third的分塊split_id3位于窗口中心window_size2使其前后各取 2 個(gè)分塊最終context_documents含 5 個(gè)文檔并按split_idx_start排序。構(gòu)造參數(shù)與 run 參數(shù)構(gòu)造函數(shù)sentence_window_retriever.pydef __init__(document_store: DocumentStore, window_size: int 3, *, source_id_meta_field: Union[str, list[str]] source_id, split_id_meta_field: str split_id, raise_on_missing_meta_fields: bool True)參數(shù)默認(rèn)值語(yǔ)義window_size3命中文檔前后各取多少個(gè)鄰近文檔如2表示取前 2 個(gè)、后 2 個(gè)必須 0_validate_window_size會(huì)對(duì) 0拋ValueError源碼第 247-250 行source_id_meta_fieldsource_id源 ID 元數(shù)據(jù)字段可傳字符串或列表傳列表時(shí)多個(gè)字段須全部匹配才視為同一來(lái)源split_id_meta_fieldsplit_id分塊位置元數(shù)據(jù)字段raise_on_missing_meta_fieldsTrue為T(mén)rue時(shí)缺少必要元數(shù)據(jù)即拋ValueError為False時(shí)跳過(guò)缺失文檔的上下文檢索但仍把原文檔放進(jìn)結(jié)果并在日志中告警run簽名component.output_types(context_windowslist[str], context_documentslist[Document]) def run(retrieved_documents: list[Document], window_size: Optional[int] None)運(yùn)行期傳入的window_size會(huì)覆蓋構(gòu)造時(shí)的值源碼第 200 行。輸出字典兩個(gè)鍵context_windows與retrieved_documents一一對(duì)應(yīng)的合并上下文文本每個(gè)字符串即一個(gè)上下文窗口context_documents原檢索文檔 其上下文文檔按split_idx_start排序。底層實(shí)現(xiàn)過(guò)濾條件構(gòu)造與去重疊合并從源碼看_retrieve_context_for_document第 265-285 行每個(gè)命中文檔的上下文獲取分三步讀取source_id可多字段與split_id任一缺失則跳過(guò)上下文檢索并告警返回(doc.content or , [doc])。用_build_filter_conditions(split_id, window_size, source_ids)構(gòu)造 AND 過(guò)濾條件第 310-322 行min_before split_id - window_size max_after split_id window_size conditions [ {field: fmeta.{split_id_meta_field}, operator: , value: min_before}, {field: fmeta.{split_id_meta_field}, operator: , value: max_after}, *source_id_filters, # 每個(gè)源字段一個(gè) {field: fmeta.{...}, operator: , value: source_id} ] return {operator: AND, conditions: conditions}調(diào)用document_store.filter_documents(filter_conditions)取回候選再由靜態(tài)方法merge_documents_text合并文本。merge_documents_text靜態(tài)方法第 119-151 行的處理細(xì)節(jié)值得注意若文檔元數(shù)據(jù)中都沒(méi)有split_idx_start則直接拼接所有文本否則按split_idx_start排序后利用max(start, last_idx_end)跳過(guò)重疊區(qū)間實(shí)現(xiàn)去重疊合并——這正是DocumentSplitter(split_overlap0)場(chǎng)景下窗口文本不重復(fù)的關(guān)鍵。run_async走完全等價(jià)的異步路徑_retrieve_context_for_document_async。兼容的 Document Store按 API 文檔聲明SentenceWindowRetriever與以下 Document Store 兼容Astra、Elasticsearch、OpenSearch、Pgvector、Pinecone、Qdrant這些 Store 均需實(shí)現(xiàn)filter_documents/filter_documents_async。AutoMergingRetriever父子層級(jí)文檔自動(dòng)合并AutoMergingRetriever基于閾值設(shè)定將命中的葉子節(jié)點(diǎn)文檔替換為其父文檔返回。它假設(shè)文檔庫(kù)中存在層級(jí)樹(shù)狀結(jié)構(gòu)葉子節(jié)點(diǎn)被索引進(jìn) Document Store父節(jié)點(diǎn)保留層級(jí)關(guān)系。設(shè)計(jì)動(dòng)機(jī)其設(shè)計(jì)動(dòng)機(jī)非常直觀一個(gè)段落被切成多個(gè)葉子分塊后若某個(gè)查詢(xún)命中了同一父節(jié)點(diǎn)下的多個(gè)分塊那么整段父文檔可能比單獨(dú)的幾個(gè)分塊更有信息量。此時(shí)返回父文檔能顯著提升下游 LLM 的上下文質(zhì)量。構(gòu)造參數(shù)與閾值校驗(yàn)def __init__(document_store: DocumentStore, threshold: float 0.5)document_store從中檢索父文檔的 Document Storethreshold決定返回父文檔還是保留子文檔的閾值。必須滿(mǎn)足0 threshold 1否則拋ValueError源碼 auto_merging_retriever.py。合并判定公式對(duì)同一父節(jié)點(diǎn)的命中子文檔數(shù)c與該父節(jié)點(diǎn)的全部子文檔數(shù)n當(dāng)c / n threshold時(shí)合并為父文檔。例如父節(jié)點(diǎn)有 3 個(gè)子文檔、命中其中 2 個(gè)時(shí)2/3 ≈ 0.67 0.5于是返回父文檔。完整示例與 HierarchicalDocumentSplitter 搭配from haystack import Document from haystack.components.preprocessors import HierarchicalDocumentSplitter from haystack.components.retrievers.auto_merging_retriever import AutoMergingRetriever from haystack.document_stores.in_memory import InMemoryDocumentStore # 創(chuàng)建 3 層層級(jí)文檔結(jié)構(gòu)父文檔有 3 個(gè)子文檔 text The sun rose early in the morning. It cast a warm glow over the trees. Birds began to sing. original_document Document(contenttext) builder HierarchicalDocumentSplitter(block_sizes[10, 3], split_overlap0, split_byword) docs builder.run([original_document])[documents] # 存儲(chǔ) level-1 父文檔并初始化檢索器 doc_store_parents InMemoryDocumentStore() for doc in docs[documents]: if doc.meta[children_ids] and doc.meta[level] 1: doc_store_parents.write_documents([doc]) retriever AutoMergingRetriever(doc_store_parents, threshold0.5) # 假設(shè)檢索到同一父節(jié)點(diǎn)下的 2 個(gè)葉子文檔 # 父節(jié)點(diǎn)有 3 個(gè)子文檔命中 2 個(gè)2/3 0.66(6)超過(guò)閾值 0.5應(yīng)返回父文檔 leaf_docs [doc for doc in docs[documents] if not doc.meta[children_ids]] docs retriever.run(leaf_docs[4:6])預(yù)期輸出返回父文檔而非兩個(gè)葉子分塊{documents: [Document(id538..), content: warm glow over the trees. Birds began to sing., meta: {block_size: 10, parent_id: 835.., children_ids: [c17..., 3ff..., 352...], level: 1, source_id: 835..., page_number: 1, split_id: 1, split_idx_start: 45})]}層級(jí)元數(shù)據(jù)約定與輸入校驗(yàn)該組件依賴(lài)HierarchicalDocumentSplitter寫(xiě)入的元數(shù)據(jù)字段見(jiàn) haystack/components/preprocessors/hierarchical_document_splitter.py__block_size當(dāng)前節(jié)點(diǎn)塊的尺寸根節(jié)點(diǎn)為 0__parent_id父文檔 ID根節(jié)點(diǎn)為None__children_ids子文檔 ID 列表葉子為空列表__level層級(jí)號(hào)根節(jié)點(diǎn)為 0逐層 1。run之前的_check_valid_documents源碼第 101-113 行會(huì)校驗(yàn)每個(gè)輸入文檔必須含__parent_id、__level、__block_size三個(gè)字段缺失即拋ValueError。注意__level與__block_size使用存在性檢查in doc.meta而非真值檢查因?yàn)楦?jié)點(diǎn)的__level0、__block_size0本身是假值。遞歸合并算法run以及等價(jià)的run_async的核心是遞歸函數(shù)_try_merge_level源碼第 141-167 行按__parent_id用defaultdict(list)對(duì)輸入文檔分組沒(méi)有父節(jié)點(diǎn)的文檔直接進(jìn)入返回列表。對(duì)每個(gè)父組通過(guò)filter_documents({field: id, operator: , value: parent_id})從 Document Store 取回父文檔要求恰好命中 1 個(gè)且父文檔必須有子節(jié)點(diǎn)否則拋ValueError。計(jì)算score len(child_docs) / len(parent_doc.meta[__children_ids])若score threshold則合并為父文檔否則子文檔原樣保留。若本輪產(chǎn)生了新的合并結(jié)果則遞歸地嘗試向更高一層合并因?yàn)楦肝臋n本身也有__parent_id直到?jīng)]有任何合并發(fā)生才返回。因此輸出可能是不同層級(jí)文檔的混合列表。兼容的 Document Store按 API 文檔聲明AutoMergingRetriever目前僅支持AstraDB、ElasticSearch、OpenSearch、PGVector、Qdrant這些 Store 需支持按 ID 過(guò)濾父文檔。序列化to_dict / from_dict 契約上述五個(gè)組件全部實(shí)現(xiàn)統(tǒng)一的序列化契約to_dict() - dict[str, Any]將組件序列化為字典。InMemory 兩個(gè)檢索器還會(huì)把filter_policy序列化為字符串值filter_policy.value便于 YAML/JSON 表示。from_dict(cls, data) - 組件實(shí)例類(lèi)方法反序列化。InMemory 兩個(gè)檢索器在from_dict中用FilterPolicy.from_str將字符串還原為枚舉后再走default_from_dict。這使得整個(gè)檢索組件可以被完整嵌入 Haystack 的 YAML 管道描述見(jiàn) haystack/marshal/yaml.py與Pipeline.loads/Pipeline.dumps流程實(shí)現(xiàn)聲明式定義、可復(fù)現(xiàn)部署。選型速查五個(gè)檢索器怎么選場(chǎng)景推薦組件關(guān)鍵理由關(guān)鍵詞/詞法匹配無(wú)需模型InMemoryBM25Retriever純 BM25 算法無(wú)需 embedding速度快、零模型依賴(lài)語(yǔ)義相似檢索已具備向量InMemoryEmbeddingRetriever接收query_embedding配合TextEmbedder/DocumentEmbedder只按元數(shù)據(jù)條件過(guò)濾FilterRetriever不做相似度計(jì)算僅執(zhí)行過(guò)濾器Comparison/Logic 語(yǔ)法檢索后需要上下文擴(kuò)展SentenceWindowRetriever基于source_id/split_id拉取鄰近分塊并去重疊合并層級(jí)分塊下避免碎片化召回AutoMergingRetriever命中同一父節(jié)點(diǎn)分塊比例超閾值時(shí)自動(dòng)升級(jí)為父文檔結(jié)語(yǔ)Haystack 2.x 的 Retrievers API 以組件化、可序列化、可異步為設(shè)計(jì)主線BM25 與 Embedding 檢索器通過(guò)run/run_async雙入口薄封裝 Document Store 的原生檢索方法FilterPolicy統(tǒng)一了運(yùn)行期與初始化過(guò)濾器的 REPLACE/MERGE 語(yǔ)義SentenceWindowRetriever與AutoMergingRetriever則分別解決了碎片化上下文與碎片化召回兩類(lèi)實(shí)際問(wèn)題。理解這些組件的參數(shù)語(yǔ)義與底層調(diào)用鏈你就能依據(jù)業(yè)務(wù)場(chǎng)景快速搭建出高質(zhì)量的 RAG 檢索階段。【免費(fèi)下載鏈接】haystackOpen-source AI orchestration framework for building context-engineered, production-ready LLM applications. Design modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation. Built for scalable agents, RAG, multimodal applications, semantic search, and conversational systems.項(xiàng)目地址: https://gitcode.com/GitHub_Trending/ha/haystack創(chuàng)作聲明：本文部分內(nèi)容由AI輔助生成（AIGC），僅供參考