Create a vector store using langchain/embedings

[cibernicola]

I am trying to create a vector store using langchain integration, but with the feature of NOT wanting to use OpenAI but a local LLM.
The embedings, I think, I'm doing them correctly, but when I try to upload them, I get errors of various types.
As in the documentation it is not clear if this step can be done or not with models other than OpenAI, I ask:

Is it possible what I am trying to do or do I have to go through ChatGPT?

[mikayelh]

hi @cibernicola - sorry you've run into an issue, we're actually pending a couple of examples with open-source models. It is certainly possible to achieve, since Deep Lake is serverless.

If you haven't yet stumbled upon our course on LangChain & Vector Databases, do check it out, especially the chapter on using GPT4all with Deep Lake.

Would you mind reporting the errors you're getting in our community slack so we can discern if these are LangChain or Deep Lake related? Thanks a lot.

[cibernicola]

I'm doing the course :D, but I haven't reached the GPT4All part and from the beginning I'm adapting everything to work with local LLM (yup I dont wanna use ChatGPT :)) and, moreover, with the most recent langchain version.

I'll keep an eye on it, although I'm using LMstudio as "engine". Let's see if I can do it.

Thanks for the course and for the answer :)

[mikayelh]

Oh, how fun! :) Thanks for letting me know - I hope you're enjoying it. Keep us posted. We've also added a course companion article last week with Cohere as an alternative to OpenAI.

[cibernicola]

I will, be sure, asking, asking, and asking :)

[cibernicola]

Well, checked, although it is not what I would have liked.
I for one love the concept of the "LLM engine" on the one hand and using it on demand on the other.

This time I use LM Studio as LLM's manager/engine, which, in turn, has the option of raising a "Local Inference Server" c

I wonder if it would be possible to use it all together: LM Studio, LangChain and deeplake.

I don't know if it is too specific. I mean, I m able to work with langchain and LM Studio, what I dont know is how to upload embeddings using deeplake

[cibernicola]

Well, as per documentation here: https://learn.activeloop.ai/courses/take/langchain/multimedia/46317979-exploring-the-role-of-langchain-s-indexes-and-retrievers

When I reach:
db.add_documents(docs)
I'm facing this problem:

Traceback (most recent call last):
  File "j:\gpt\llm_tests\5-Keeping Knowledge Organized with Indexes\tmp.py", line 97, in <module>
    test()
  File "j:\gpt\llm_tests\5-Keeping Knowledge Organized with Indexes\tmp.py", line 76, in test
    db.add_documents(docs_from_file)
  File "C:\Python310\lib\site-packages\langchain\vectorstores\base.py", line 104, in add_documents
    return self.add_texts(texts, metadatas, **kwargs)
  File "C:\Python310\lib\site-packages\langchain\vectorstores\deeplake.py", line 224, in add_texts
    embedding_function=self._embedding_function.embed_documents,  # type: ignore
AttributeError: 'list' object has no attribute 'embed_documents'

The content of "docs" looks like:

[Document(page_content='Google opens up its AI language model PaLM to challenge OpenAI and GPT-3\n    Google is offering developers access to one of its most advanced AI language models: PaLM.\n    The search giant is launching an API for PaLM alongside a number of AI enterprise tools\n    it says will help businesses “generate text, images, code, videos, audio, and more from\n    simple natural language prompts.”', metadata={'source': 'my_file.txt'}), Document(page_content='PaLM is a large language model, or LLM, similar to the GPT series created by OpenAI or\n    Meta’s LLaMA family of models. Google first announced PaLM in April 2022. Like other 
LLMs,\n    PaLM is a flexible system that can potentially carry out all sorts of text generation and\n    editing tasks. You could train PaLM to be a conversational chatbot like ChatGPT, for\n    example, or you could use it for tasks like summarizing text or even writing code.\n    (It’s similar to features Google also announced today for its Workspace apps like Google\n    Docs and Gmail.)', metadata={'source': 'my_file.txt'})]

I'm barely lost, don't know where the problem is :/

[adolkhan]

Hey @cibernicola! Thanks for reporting! We are taking a look, will get back to as soon as I will have some results

[cibernicola]

Hello!

Just updated deeplake to deeplake-3.6.19

And now, finally, it works!!!

Edit: still working on it :S

[adolkhan]

Hey @cibernicola! Great it worked! Do you have any more questions?

[cibernicola]

Tons, but I want to get to the end, since there are parts that depend on langchain, others on deeplake and most of it on me :)

Let's continue!