Langchain integration

[pseudotensor]

• basic isolated demo with OpenAI or HF/sentence_transformers
• use better embedding and LLM (ours) to see if comparable to OpenAI
• incorporate into chatbot with PDF upload
• If well-chunked, extra chunking is not good idea, except to keep below (perhaps) slightly larger limit than chunk_size
• Add pre-dB building for in-memory for now, not when user asks for response
• Avoid re-downloading github etc. each time
• Make chroma work, need to understand how to avoid dup docs and why fails when just load (missing items)
• How to choose cut_distanct?
• Context not added from chat when find sources. Maybe that's ok-ish, but need to do better.
• Add docs: https://huggingface.co/databricks/dolly-v2-12b#langchain-usage
• https://huggingface.co/datasets/Cohere/wikipedia-22-12/viewer/en/train
• Fix context so doesn't do hard chop, do at human-bot instead.

Future PRs:

• Make URL getting work
• Can ground report more by showing matching words as well as matching n-grams from actual sources
• Add ability to do new separate chat and go back to other chats
• Add ability to download chat(s).
• Consume url from chat itself and do look up
• https://github.com/h2oai/makersaurus
• Make All just mean look at all dbs
• Allow any github link to ingest if not on HF. If on HF, default to h2oGPT and DAI docs.
• Unsure how to work around [Bug]: Failed to Deploy in HuggingFace Spaces chroma-core/chroma#412 (comment)
• Good or Bad for Flag, and Compare should be which is better

[pseudotensor]

Using sentence transformer + our 6.9B model, gives consistent answers.

query: Which config.toml enables pytorch for NLP?
answer: The following config.toml file enables PyTorch for NLP:

```
[nlp]
enable_pytorch_nlp_model = true
enable_pytorch_nlp_transformer = true
```

This file can be found in the `config` directory of the DriverlessAI documentation.
sources: ['nlp_settings.rst', 'nlp.rst', 'nlp_settings.rst', 'nlp_settings.rst']

Similar (previous one with sources mentioned in prompt but not actually present anymore) direct prompt in llama does well too:

[pseudotensor]

============================================================================================= short test summary info =============================================================================================
FAILED tests/test_manual_test.py::test_chat_context - NotImplementedError: MANUAL TEST FOR NOW
=================================================================== 1 failed, 13 passed, 7 skipped, 1 xpassed, 9 warnings in 364.21s (0:06:04) ====================================================================
(h2ollm) jon@pseudotensor:~/h2o-llm$ 

and tested manual test manually.