Using as a Library

Using as a Library in Your Own App

Basic integration pattern

from multi_turboquant import CacheConfig, CacheMethod, get_preset
from multi_turboquant.integration import get_llamacpp_command
from multi_turboquant.hardware import detect_gpus

# Detect hardware
gpus = detect_gpus()

# Get a preset (or let the user choose)
config = get_preset("balanced")

# Generate the launch command
cmd = get_llamacpp_command(
    config,
    model_path="/opt/models/model.gguf",
    port=8080,
)

# Start the server with subprocess
import subprocess
proc = subprocess.Popen(cmd)

UI dropdown population

from multi_turboquant.integration import BridgeAdapter

adapter = BridgeAdapter(config)
options = adapter.get_ui_options()
# Returns list of dicts with value, label, group, needs_calibration, description
# Ready to populate a <select> element

Real-time capacity display

from multi_turboquant import plan_agents
from multi_turboquant.hardware import detect_gpus

def on_config_change(model_size, agents, context):
    """Called when user adjusts sliders in the UI."""
    gpus = [g.to_planner_dict() for g in detect_gpus()]
    result = plan_agents(
        gpus=gpus,
        model_params_b=model_size,
        desired_agents=agents,
        desired_context=context,
    )
    return result.to_dict()  # JSON-serializable for the frontend

Auto-calibration before service start

from multi_turboquant.calibration import auto_calibrate

# Call before starting inference with TurboQuant
results = auto_calibrate(config, "/opt/models/MyModel")
if results:
    print(f"Generated: {results}")

Full example

See multi_turboquant/integration/example_app_integration.py for a complete integration example showing UI dropdown, capacity planning, auto-calibration, and command generation.