When using embedding, the Chinese reply will be incomplete

[lingfengchencn]

System Info

Mac vs code python :Python 3.10.11

Who can help?

No response

Information

• [ ] The official example notebooks/scripts
• [X] My own modified scripts

Related Components

• [ ] LLMs/Chat Models
• [X] Embedding Models
• [ ] Prompts / Prompt Templates / Prompt Selectors
• [X] Output Parsers
• [ ] Document Loaders
• [ ] Vector Stores / Retrievers
• [ ] Memory
• [ ] Agents / Agent Executors
• [ ] Tools / Toolkits
• [ ] Chains
• [ ] Callbacks/Tracing
• [ ] Async

Reproduction

1. embedding some text into Chroma
2. query and run load_qa_chain with OpenAI

docs = docsearch.similarity_search(query="some txt",k=2)
llm = OpenAI(
        streaming=True, 
        callbacks=[StreamingStdOutCallbackHandler()],
        temperature=0.1)
chain = load_qa_chain(llm=llm,chain_type="stuff",verbose=True)
result = chain.run(input_documents=docs,question=query,return_only_outputs=True)

3. The result in Chinese keeps 127 ~131 words, English will finish the whole sentence.

example:

我们***是一家专注于*****机构，近些年来，我们的学员人数突破****，遍布全国***个城市，海外**个国家，这自然是我们家长对于****最好的认可。我们深知宝贝一开始有兴趣，后来因为各种的枯燥变得不愿意学了，因此，我们采用三方合作配合的模式，即家长

我们***是一家专注于*****机构，近些年来，我们的学员人数突破****，遍布全国***个城市，海外**个国家，这自然是我们家长对于***最好的认可。我们深知宝贝一开始有兴趣，后来因为各种的枯燥变得不愿意学了的顾虑，因此我们采用了一种科学的学习模式

Expected behavior

I think this was posted while working on characters, looking forward to a fix.

[qazs]

Same issue here

[qazs]

Found the solution, you can add a max_tokens parameters when initializing OpenAI like this:

llm = OpenAI(temperature=0, max_tokens=2048)

[lingfengchencn]

tks， I upgrade langchain ,and it's gone..

[chujian]

你还是把日志打印出来，看看检索到的2个片段是否是你要的答案，另外文档的分割也很重要。调整你文档的段落，重新向量化以后看看