Skip to content

Latest commit

 

History

History
84 lines (55 loc) · 2.32 KB

readme.md

File metadata and controls

84 lines (55 loc) · 2.32 KB

Private GPT

What is private GPT

How to build and run the project

building the project

  • create conda environment and activate it 
    conda create -p venv python --y
conda info --envs
conda activate <venv>
  • install the requirements, it takes some time (7-10 minutes) 
    pip install -r requirements.txt
  • download model from huggingface 
    https://huggingface.co/vicuna/ggml-vicuna-13b-1.1/tree/main

System Requirements

Flow

  • ingest.py

    • This script processes the documents stored in the source_docs folder
    • while processing the document, this script loads only the new documents in the folder
    • langchain document loaders are used to load the files based on their extentions
    • landchain text splitter is used to recursively split the text to chunk size and overlap
    • these texts are then stored in the vector db using the huggingface embeddings
    • this process converts the documents to embeddings which the langchain models understands

Terminologies

  • Embedding
    An embedding is a numerical representation of a piece of information, for example, text, documents, images, audio, etc. The representation captures the semantic meaning of what is being embedded

Reasonings

  • Why use all-MiniLM-L6-v2 embedding models ? https://www.sbert.net/docs/pretrained_models.html
    As per above link, all-MiniLM-L6-v2 is 5 times faster but still provides good quality. We can use other other models as per our requirements.

References

fixing

pip install certifi
python -m certifi
ln -s /opt/anaconda3/envs/privategpt/lib/python3.12/site-packages/certifi/cacert.pem cert.pem
rehash

https://stackoverflow.com/a/75111104

https://discuss.huggingface.co/t/using-huggingface-embeddings-completely-locally/70837

download manually : https://www.kaggle.com/datasets/shinomoriaoshi/sentencetransformersallminilml6v2?resource=download

UKPLab/sentence-transformers#1462

curl --insecure -O https://public.ukp.informatik.tu-darmstadt.de/reimers/sentence-transformers/v0.2/distilbert-base-nli-stsb-mean-tokens.zip

unzip distilbert-base-nli-stsb-mean-tokens.zip -d ./sent_bert

pip install –trusted-host pypi.org
–trusted-host huggingface.co \