如何查看LangChain中Chroma存储的文档Embedding?为何返回None?
Chroma搭配LangChain+OpenAI Embeddings时无法查看向量embeddings字段的问题
执行db.get()后,返回结果中的embeddings字段始终为None,但文档内容、元数据等其他内容均正常显示。
已做的验证测试
- 单独验证embeddings组件功能正常,执行以下代码返回向量长度为1536:
len(embeddings.embed_documents(["EMBED THIS"])[0]) >> 1536
- 本地
./data目录下存在chroma-embeddings.parquet文件
官方示例代码复现问题
即使使用官方示例代码创建Chroma实例,执行new_db.get()后,embeddings字段仍为None:
# Import Document class from langchain.docstore.document import Document # Initial document content and id initial_content = "This is an initial document content" document_id = "doc1" # Create an instance of Document with initial content and metadata original_doc = Document(page_content=initial_content, metadata={"page": "0"}) # Initialize a Chroma instance with the original document new_db = Chroma.from_documents( collection_name="test_collection", documents=[original_doc], embedding=OpenAIEmbeddings(), # using the same embeddings as before ids=[document_id], )
我的核心代码
for db_collection_name in tqdm(["class1-sub2-chap3", "class2-sub3-chap4"]): documents = [] doc_ids = [] for doc_index in range(3): cl, sub, chap = db_collection_name.split("-") content = f"This is {db_collection_name}-doc{doc_index}" doc = Document(page_content=content, metadata={"chunk_num": doc_index, "chapter":chap, "class":cl, "subject":sub}) documents.append(doc) doc_ids.append(str(doc_index)) # Initialize a Chroma instance with the original document db = Chroma.from_documents( collection_name=db_collection_name, documents=documents, ids=doc_ids, embedding=embeddings, persist_directory="./data") db.persist()
执行db.get()的返回结果示例
{'ids': ['0', '1', '2'], 'embeddings': None, 'documents': ['This is class1-sub2-chap3-doc0', 'This is class1-sub2-chap3-doc1', 'This is class1-sub2-chap3-doc2'], 'metadatas': [{'chunk_num': 0, 'chapter': 'chap3', 'class': 'class1', 'subject': 'sub2'}, {'chunk_num': 1, 'chapter': 'chap3', 'class': 'class1', 'subject': 'sub2'}, {'chunk_num': 2, 'chapter': 'chap3', 'class': 'class1', 'subject': 'sub2'}]}
内容的提问来源于stack exchange,提问作者Deshwal
相关产品推荐
相关产品推荐

