split_big_quants.sh

#!/bin/bash

# Path to llama-gguf-split
SPLIT_TOOL=~/git/llama.cpp/build/bin/llama-gguf-split

# Define the 40GB limit in bytes
LIMIT_BYTES=42949672960 # 40GB (1024^3 * 40)

echo "Starting GGUF splitting for files > 40GB..."
echo "----------------------------------------"

# Find all .gguf files in the current directory
find . -maxdepth 1 -name "*.gguf" -print0 | while IFS= read -r -d $'\0' FILE;
do
    # Remove leading ./ for cleaner filename display
    FILE_NAME=${FILE#./}

    # Skip Ina-v11.1-F16.gguf (original F16 file)
    if [[ "$FILE_NAME" == "Ina-v11.1-F16.gguf" ]]; then
        echo "Skipping original F16 file: $FILE_NAME"
        continue
    fi

    # Get file size
    FILE_SIZE=$(stat -f%z "$FILE_NAME")

    if [ "$FILE_SIZE" -gt "$LIMIT_BYTES" ]; then
        echo "Processing: $FILE_NAME (Size: $(numfmt --to=iec-i --suffix=B "$FILE_SIZE"))"
        
        # Create directory name (remove .gguf extension)
        DIR_NAME="${FILE_NAME%.gguf}"
        mkdir -p "$DIR_NAME"

        echo "  Created directory: $DIR_NAME"
        echo "  Splitting '$FILE_NAME' into '$DIR_NAME/'..."

        # Change to the new directory to run the split command
        pushd "$DIR_NAME" > /dev/null

        # Run split command: GGUF_IN is relative path, GGUF_OUT is basename as prefix
        "$SPLIT_TOOL" --split-max-size 40G "../$FILE_NAME" "$(basename "$FILE_NAME" .gguf)"
        
        SPLIT_EXIT=$?

        # Change back to original directory
        popd > /dev/null

        if [ $SPLIT_EXIT -eq 0 ]; then
            echo "✅ Split successful. Removing original large file: $FILE_NAME"
            rm "$FILE_NAME"
        else
            echo "❌ Splitting failed for $FILE_NAME (Error code: $SPLIT_EXIT). Keeping original file."
        fi
    else
        echo "Skipping $FILE_NAME (Size: $(numfmt --to=iec-i --suffix=B "$FILE_SIZE")) - smaller than 40GB."
    fi
    echo "----------------------------------------"
done

echo "GGUF splitting complete."