llama2.sh

#!/bin/bash

# Default values
MODEL="llama-2-13b-chat.ggmlv3.q4_0.bin"
THREADS=8
NGL=1
COLOR="--color"
C_FLAG=2048
TEMP=0.7
REPEAT_PENALTY=1.1
N_FLAG=-1

# Function to display help menu
function display_help() {
    echo "Usage: $0 [options]"
    echo "  Options:"
    echo "  --model           specify the model name (default: llama-2-13b-chat.ggmlv3.q4_0.bin)"
    echo "  --threads         specify the number of threads (default: 8)"
    echo "  --ngl             specify the ngl (default: 1)"
    echo "  --color           specify the color (default: --color)"
    echo "  -c                specify the c flag (default: 2048)"
    echo "  --temp            specify the temp (default: 0.7)"
    echo "  --repeat_penalty  specify the repeat_penalty (default: 1.1)"
    echo "  -n                specify the n flag (default: -1)"
    echo "  -h, --help        display this help and exit"
    exit 1
}

# Parse command line arguments
while [[ $# -gt 0 ]]; do
    key="$1"

    case $key in
        --model)
        MODEL="$2"
        shift # past argument
        shift # past value
        ;;
        --threads)
        THREADS="$2"
        shift
        shift
        ;;
        --ngl)
        NGL="$2"
        shift
        shift
        ;;
        --color)
        COLOR="$2"
        shift
        shift
        ;;
        -c)
        C_FLAG="$2"
        shift
        shift
        ;;
        --temp)
        TEMP="$2"
        shift
        shift
        ;;
        --repeat_penalty)
        REPEAT_PENALTY="$2"
        shift
        shift
        ;;
        -n)
        N_FLAG="$2"
        shift
        shift
        ;;
        -h|--help)
        display_help
        shift # past argument
        ;;
        *)    # unknown option
        shift # past argument
        ;;
    esac
done

REPO_URL="https://github.com/ggerganov/llama.cpp.git"
MODEL_URL="https://huggingface.co/TheBloke/Llama-2-13B-chat-GGML/resolve/main/${MODEL}"

# Check if llama.cpp directory already exists
if [ -d "llama.cpp" ]; then
    echo "Directory llama.cpp already exists. Pulling the latest changes..."
    cd llama.cpp
    git pull
else
    echo "Cloning llama.cpp..."
    git clone ${REPO_URL}
    cd llama.cpp
fi

# Build it
echo "Building the project..."
LLAMA_METAL=1 make

# Check if model already exists
if [ -f "${MODEL}" ]; then
    echo "Model ${MODEL} already exists."
else
    echo "Downloading the model..."
    wget -q --show-progress ${MODEL_URL}
fi

# Run the program
while true; do
    read -p "Enter your prompt (or 'exit' to quit): " PROMPT
    if [ "$PROMPT" == "exit" ]; then
        echo "Exiting the program."
        break
    fi
    ./main \
        -t ${THREADS} \
        -ngl ${NGL} \
        -m ${MODEL} \
        ${COLOR} \
        -c ${C_FLAG} \
        --temp ${TEMP} \
        --repeat_penalty ${REPEAT_PENALTY} \
        -n ${N_FLAG} \
        -p "${PROMPT}"
done

why wget over curl?