install-text-generation-inference

#!/bin/sh

if test ! -f fasteval ; then
    echo 'ERROR: Please run this command from the `FastEval` directory'
    exit
fi

git clone --depth 1 https://github.com/huggingface/text-generation-inference.git
cd text-generation-inference

# We will use *a second virtual environment* specifically for `text-generation-inference`. This is to avoid package conflicts.
python3.10 -m venv .venv
. .venv/bin/activate

# This is needed for parallel builds which are significantly faster.
pip install ninja

BUILD_EXTENSIONS=True make install

cd server
make install-flash-attention
make install-flash-attention-v2
make install-vllm # Installs *a part* of vLLM used in text-generation-inference. This is not the full vLLM that we already installed.