#!/usr/bin/env python3
"""
Sixpert K2 - OpenAI-Compatible API Server
==========================================
Runs Sixpert K2 as an OpenAI-compatible API server using llama-cpp-python.

The MoE architecture means K2 activates only ~1.2B parameters per token,
making it significantly faster than dense models of similar total parameter count.

Usage:
    pip install llama-cpp-python
    python api_server.py

Then use with any OpenAI-compatible client:
    curl http://localhost:8000/v1/chat/completions \
      -H "Content-Type: application/json" \
      -d '{
        "model": "sixpert-k2",
        "messages": [{"role": "user", "content": "Explain quantum computing deeply"}]
      }'
"""

import argparse
from llama_cpp.server.app import create_app
from llama_cpp import Llama


def main():
    parser = argparse.ArgumentParser(description="Sixpert K2 API Server")
    parser.add_argument(
        "--model", type=str, default="SixpertK2.gguf", help="Path to GGUF model"
    )
    parser.add_argument("--host", type=str, default="0.0.0.0", help="Server host")
    parser.add_argument("--port", type=int, default=8000, help="Server port")
    parser.add_argument("--gpu-layers", type=int, default=-1, help="GPU layers (-1 = all)")
    parser.add_argument("--threads", type=int, default=8, help="CPU threads")

    args = parser.parse_args()

    print("=" * 60)
    print("  Sixpert K2 API Server")
    print("  Deep Reasoning Engine (MoE)")
    print("  Total: ~8.9B params | Active: ~1.2B per token")
    print("=" * 60)

    llm = Llama(
        model_path=args.model,
        n_ctx=131072,
        n_gpu_layers=args.gpu_layers,
        n_threads=args.threads,
        verbose=False,
    )

    app = create_app(llm)

    print(f"\n  Server running at http://{args.host}:{args.port}")
    print(f"  Model: sixpert-k2")
    print(f"  Endpoint: POST /v1/chat/completions")
    print(f"  Endpoint: POST /v1/completions")
    print("=" * 60)

    import uvicorn
    uvicorn.run(app, host=args.host, port=args.port)


if __name__ == "__main__":
    main()