Felipe97/llama-cpp-compiled

1#!/usr/bin/env bash2 3set -e4 5CONVERTED_MODEL="${1:-"$CONVERTED_MODEL"}"6QUANTIZED_TYPE="${2:-"$QUANTIZED_TYPE"}"7TOKEN_EMBD_TYPE="${3:-"${TOKEN_EMBD_TYPE}"}"8OUTPUT_TYPE="${4:-"${OUTPUT_TYPE}"}"9BUILD_DIR="${5:-"$BUILD_DIR"}"10QUANTIZED_MODEL=$CONVERTED_MODEL11 12# Final check if we have a model path13if [ -z "$CONVERTED_MODEL" ]; then14    echo "Error: Model path must be provided either as:" >&215    echo "  1. Command line argument" >&216    echo "  2. CONVERTED_MODEL environment variable" >&217    exit 118fi19 20if [ -z "$QUANTIZED_TYPE" ]; then21    echo "Error: QUANTIZED_TYPE is required" >&222    exit 123fi24 25echo $CONVERTED_MODEL26 27# Process the quantized model filename28if [[ "$QUANTIZED_MODEL" == *.gguf ]]; then29    # Remove .gguf suffix, add quantized type, then add .gguf back30    BASE_NAME="${QUANTIZED_MODEL%.gguf}"31    QUANTIZED_MODEL="${BASE_NAME}-${QUANTIZED_TYPE}.gguf"32else33    echo "Error: QUANTIZED_MODEL must end with .gguf extension" >&234    exit 135fi36 37if [ -z "$BUILD_DIR" ]; then38    BUILD_DIR="../../build"39fi40 41cmake --build $BUILD_DIR --target llama-quantize -j842 43echo $TOKEN_EMBD_TYPE44echo $OUTPUT_TYPE45 46CMD_ARGS=("${BUILD_DIR}/bin/llama-quantize")47[[ -n "$TOKEN_EMBD_TYPE" ]] && CMD_ARGS+=("--token-embedding-type" "$TOKEN_EMBD_TYPE")48[[ -n "$OUTPUT_TYPE" ]]     && CMD_ARGS+=("--output-tensor-type" "$OUTPUT_TYPE")49CMD_ARGS+=("$CONVERTED_MODEL" "$QUANTIZED_MODEL" "$QUANTIZED_TYPE")50 51"${CMD_ARGS[@]}"52 53echo "Quantized model saved to: $QUANTIZED_MODEL"54