CyberRanger/colab_notebooks/merge_v19_to_ollama.py

#!/usr/bin/env python3
"""
CyberRanger V19 Adapter Merge Script
Merges QLoRA adapter with base model and prepares for Ollama

Usage:
    python3 merge_v19_to_ollama.py

Requirements:
    pip install torch transformers peft accelerate
"""

import os
import sys
import shutil
from pathlib import Path

# Paths
ADAPTER_PATH = "/Users/ranger/.ranger-memory/NCI_MSc/Year_2/AI_ML_Cybersecurity/Colab_Forge_V5/CyberRanger_V19_Adapters"
BASE_MODEL = "HuggingFaceTB/SmolLM2-1.7B-Instruct"  # or SmolLM2-1.7B if you used base
OUTPUT_DIR = "/Users/ranger/.ranger-memory/code/qbrain/merged_v19"
MODELFILE_PATH = "/Users/ranger/.ranger-memory/code/qbrain/Modelfile.v19-mesh"

def check_dependencies():
    """Check if required packages are installed"""
    try:
        import torch
        import transformers
        import peft
        print("✅ All dependencies installed")
        print(f"   PyTorch: {torch.__version__}")
        print(f"   Transformers: {transformers.__version__}")
        print(f"   PEFT: {peft.__version__}")
        return True
    except ImportError as e:
        print(f"❌ Missing dependency: {e}")
        print("\nInstall with:")
        print("pip install torch transformers peft accelerate")
        return False

def merge_adapter():
    """Merge QLoRA adapter with base model"""
    from transformers import AutoModelForCausalLM, AutoTokenizer
    from peft import PeftModel
    import torch

    print("\n🔄 STEP 1: Loading base model...")
    print(f"   Base: {BASE_MODEL}")

    # Load base model
    base_model = AutoModelForCausalLM.from_pretrained(
        BASE_MODEL,
        torch_dtype=torch.float16,
        device_map="auto",
        trust_remote_code=True
    )

    tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)

    print("✅ Base model loaded")

    print("\n🔄 STEP 2: Loading V19 adapter...")
    print(f"   Adapter: {ADAPTER_PATH}")

    # Load adapter
    model = PeftModel.from_pretrained(base_model, ADAPTER_PATH)
    print("✅ Adapter loaded")

    print("\n🔄 STEP 3: Merging adapter into base model...")
    merged_model = model.merge_and_unload()
    print("✅ Merge complete!")

    print(f"\n🔄 STEP 4: Saving merged model to {OUTPUT_DIR}...")
    os.makedirs(OUTPUT_DIR, exist_ok=True)
    merged_model.save_pretrained(OUTPUT_DIR, safe_serialization=True)
    tokenizer.save_pretrained(OUTPUT_DIR)
    print("✅ Merged model saved!")

    # Copy the Modelfile system prompt
    modelfile_dest = os.path.join(OUTPUT_DIR, "Modelfile")
    if os.path.exists(MODELFILE_PATH):
        shutil.copy(MODELFILE_PATH, modelfile_dest)
        print(f"✅ Modelfile copied to {modelfile_dest}")

    return OUTPUT_DIR

def print_next_steps(output_dir):
    """Print instructions for GGUF conversion and Ollama"""
    print("\n" + "="*60)
    print("🎖️ V19 ADAPTER MERGE COMPLETE!")
    print("="*60)
    print(f"\nMerged model saved to: {output_dir}")
    print("\n📋 NEXT STEPS:")
    print("-" * 40)
    print("""
1. Clone llama.cpp (if not already done):
   git clone https://github.com/ggerganov/llama.cpp
   cd llama.cpp
   pip install -r requirements.txt

2. Convert to GGUF:
   python3 llama.cpp/convert_hf_to_gguf.py \\
       {output_dir} \\
       --outfile rangerbot-v19.gguf \\
       --outtype f16

3. (Optional) Quantize for smaller size:
   ./llama.cpp/llama-quantize rangerbot-v19.gguf rangerbot-v19-q4.gguf q4_k_m

4. Create Modelfile for GGUF:
   echo 'FROM ./rangerbot-v19.gguf' > Modelfile.gguf
   # Then add the SYSTEM prompt from Modelfile.v19-mesh

5. Create Ollama model:
   ollama create rangerbot:v19 -f Modelfile.gguf

6. Test it:
   ollama run rangerbot:v19
""".format(output_dir=output_dir))
    print("="*60)
    print("Rangers lead the way! 🎖️💥🧠")
    print("="*60)

def main():
    print("="*60)
    print("🎖️ CYBERRANGER V19 ADAPTER MERGE SCRIPT")
    print("="*60)

    # Check dependencies
    if not check_dependencies():
        sys.exit(1)

    # Check adapter exists
    if not os.path.exists(os.path.join(ADAPTER_PATH, "adapter_model.safetensors")):
        print(f"❌ Adapter not found at {ADAPTER_PATH}")
        sys.exit(1)

    print(f"\n✅ V19 adapter found: {ADAPTER_PATH}")

    # Merge
    try:
        output_dir = merge_adapter()
        print_next_steps(output_dir)
    except Exception as e:
        print(f"\n❌ Error during merge: {e}")
        import traceback
        traceback.print_exc()
        sys.exit(1)

if __name__ == "__main__":
    main()