35 lines
935 B
Bash
Executable File
35 lines
935 B
Bash
Executable File
#!/bin/bash
|
|
|
|
set -e
|
|
|
|
CONVERTED_MODEL="${1:-"$CONVERTED_MODEL"}"
|
|
QUANTIZED_TYPE="${2:-"$QUANTIZED_TYPE"}"
|
|
QUANTIZED_MODEL=$CONVERTED_MODEL
|
|
|
|
# Final check if we have a model path
|
|
if [ -z "$CONVERTED_MODEL" ]; then
|
|
echo "Error: Model path must be provided either as:" >&2
|
|
echo " 1. Command line argument" >&2
|
|
echo " 2. CONVERTED_MODEL environment variable" >&2
|
|
exit 1
|
|
fi
|
|
|
|
echo $CONVERTED_MODEL
|
|
|
|
# Process the quantized model filename
|
|
if [[ "$QUANTIZED_MODEL" == *.gguf ]]; then
|
|
# Remove .gguf suffix, add quantized type, then add .gguf back
|
|
BASE_NAME="${QUANTIZED_MODEL%.gguf}"
|
|
QUANTIZED_MODEL="${BASE_NAME}-${QUANTIZED_TYPE}.gguf"
|
|
else
|
|
echo "Error: QUANTIZED_MODEL must end with .gguf extension" >&2
|
|
exit 1
|
|
fi
|
|
|
|
|
|
cmake --build ../../build --target llama-quantize -j8
|
|
|
|
../../build/bin/llama-quantize $CONVERTED_MODEL $QUANTIZED_MODEL $QUANTIZED_TYPE
|
|
|
|
echo "Quantized model saved to: $QUANTIZED_MODEL"
|