Skip to content

Latest commit

 

History

History
30 lines (20 loc) · 1005 Bytes

README_RWKV.md

File metadata and controls

30 lines (20 loc) · 1005 Bytes
Raw

Installation and Setup

  • Install Miniconda

  • Create a conda environment

conda create -n llama-index python=3.11.3
conda activate llama-index
  • Install g++ and CUDA
mamba install -c nvidia -c conda-forge cxx-compiler==1.1.3 cuda cuda-toolkit

  • Install the Python package with pip install -r requirements.txt

  • Download a RWKV model and place it in your desired directory

    mkdir models/rwkv-4-raven
wget xxx -O ./models/rwkv-4-raven/RWKV-4-Raven-14B-v11x-Eng99-Other1-20230501-ctx8192.pth
wget https://raw.githubusercontent.com/BlinkDL/ChatRWKV/main/20B_tokenizer.json -O ./models/rwkv-4-raven/20B_tokenizer.json

  • Download the tokens file

More details on the model and tokenizer can be found here.