Felipe97/llama-cpp-compiled

1#!/usr/bin/env bash2 3set -e4 5MODEL_PATH="${1:-"$MODEL_PATH"}"6MODEL_NAME="${2:-$(basename "$MODEL_PATH")}"7 8CONVERTED_MODEL_PATH="${1:-"$CONVERTED_MODEL"}"9CONVERTED_MODEL_NAME="${2:-$(basename "$CONVERTED_MODEL_PATH" ".gguf")}"10 11if [ -t 0 ]; then12    CPP_EMBEDDINGS="data/llamacpp-${CONVERTED_MODEL_NAME}-embeddings.bin"13else14    # Process piped JSON data and convert to binary (matching logits.cpp format)15    TEMP_FILE=$(mktemp /tmp/tmp.XXXXXX.binn)16    python3 -c "17import json18import sys19import struct20 21data = json.load(sys.stdin)22 23# Flatten all embeddings completely24flattened = []25for item in data:26    embedding = item['embedding']27    for token_embedding in embedding:28        flattened.extend(token_embedding)29 30print(f'Total embedding values: {len(flattened)}', file=sys.stderr)31 32# Write as binary floats - matches logitc.cpp fwrite format33with open('$TEMP_FILE', 'wb') as f:34    for value in flattened:35        f.write(struct.pack('f', value))36"37    CPP_EMBEDDINGS="$TEMP_FILE"38    trap "rm -f $TEMP_FILE" EXIT39fi40 41python scripts/utils/semantic_check.py --model-path $MODEL_PATH \42    --python-embeddings data/pytorch-${MODEL_NAME}-embeddings.bin \43    --cpp-embeddings $CPP_EMBEDDINGS \44    --prompt "Hello world today" \45    --causal46 47