w-wuxian/nanika

MistalAI playground

Testing langchain rag

Get branch ollama-rag from Nanika repo

The branch focus the use of ollama and langchain to RAG from your documents. (see list of supported file extensions at the end)

cd $HOME
git clone -b ollama-rag --single-branch [email protected]:W-Wuxian/NANIKA.git

Install ollama

After a successful installation run:

ollama pull nomic-embed-text 
ollama pull phi3
ollama list

nomic-embed-text is mandatory but phi3 can be replaced with any model name at (ollama.com/library)[https://ollama.com/library]

Activate langchain dependencies:

After installing ollama materials you need to do the following:

conda env create -f langchain_rag_env.yml
conda activate langchain_rag_env
pip install "unstructured[all-docs]"
pip install chromadb langchain-text-splitters
conda install conda-forge::pytesseract
conda install conda-forge::tesseract

Alternative using Python-venv

python -m venv langchain_rag_venv
pip install --upgrade unstructured langchain "unstructured[all-docs]"
pip install --upgrade chromadb langchain-text-splitters
pip install --upgrade pytesseract
pip install --upgrade tesseract

Running the code:

Once ollama and langchain stuff are done (see previous sections) you can use RAG. Here is two python scripts nanika.py to do so.

Creating a database and QA loop

The nanika.py script is used to create a database from your documents as follow:

python nanika.py --help
options or long_options are:
-m or --model_name model name
-e or --embedding_name embedding name
-i or --inputdocs_path  path given between " " to folders or files to be used at RAG step
-v or --vdb_path vector data base path
-c or --collection_name collection name
-r or --reuse reuse previous vdb and collection
-d or --display-doc whether or not to display given documents

So for example using phi3 llm model, with nomic-embed-text as an embedding model to create a database from my documents at /path/to/my/folder/ one can use the following command:

python nanika.py -m phi3 -e nomic-embed-text -i "/path/to/my/folder1 /path/to/my/folder2 /path/to/my/file1"

In order to run several database we need to specify the database storing location via -v and the collection name via -c, as follow:

python nanika.py -m phi3 -e nomic-embed-text -i /path/to/my/folder1/ -v ./database1 -c collection1
python nanika.py -m phi3 -e nomic-embed-text -i /path/to/my/folder2/ -v ./database2 -c collection2

The nanika.py script will also ask you to enter questions (RAG), to end this phase enter q or quit.

Reusing a database and QA loop

To reuse a database you need the corresponding -v and -c and run the nanika.py script with -r True as follow:

python nanika.py -m phi3 -e nomic-embed-text -v ./database1 -c collection1 -r True
python nanika.py -m phi3 -e nomic-embed-text -v ./database2 -c collection2 -r True

File extension coverage

file extension	Coverage
pdf	✔️
txt	✔️
py	✔️
png	✔️
jpg	✔️
xlsx	✔️
xls	✔️
odt	✔️
csv	✔️
pptx	✔️
md	✔️
org	✔️