To work with those examples you will need a Hugging Face Text Generation Inference server deployed, and a model served.
You can either install Langchain and its dependencies in your workbench (pip install langchain), or you can directly use a pre-built custom Workbench image that comes with everything needed: quay.io/opendatahub-contrib/workbench-images:cuda-jupyter-langchain-c9s-py311_2023c_latest.
Ref: How to import a custom notebook image
If you want to create your own custom image, with a different IDE like VSCode for example, you will find instructions here.
- Langchain-Caikit-Basic.ipynb: Basic example on using Langchain to query Llama2 served through Caikit+TGIS.
- Langchain-Caikit-Prompt-memory.ipynb: More advanced example on using Langchain to query Llama2 served through Caikit+TGIS, with custom prompt and conversation memory buffer.
- Langchain-HFTGI-Basic.ipynb: Basic example on using Langchain to query Llama2 served through Hugging Face TGI.
- Langchain-HFTGI-Prompt-memory.ipynb: More advanced example on using Langchain to query Llama2 served through Hugging Face TGI, with custom prompt and conversation memory buffer.
- RAG_with_sources_Langchain-HFTGI.ipynb: Retrieval Augmented Generation (ask questions about documents) using Langchain with Llama2 served through Hugging Face TGI.
- RAG_with_sources_Langchain-Caikit.ipynb: Retrieval Augmented Generation (ask questions about documents) using Langchain with Llama2 served through Caikit+TGIS.