app.py

from fastapi import FastAPI, Request
from llama_cpp import Llama
from huggingface_hub import hf_hub_download
import os
import platform
import psutil
import multiprocessing
import time

app = FastAPI()

# === Model Config ===
REPO_ID = "TheBloke/phi-2-GGUF"
FILENAME = "phi-2.Q3_K_M.gguf"
MODEL_DIR = "models"
MODEL_PATH = os.path.join(MODEL_DIR, FILENAME)

# === Download if model not available ===
if not os.path.exists(MODEL_PATH):
    print(f"⬇️ Downloading {FILENAME} from Hugging Face...")
    model_path = hf_hub_download(
        repo_id=REPO_ID,
        filename=FILENAME,
        cache_dir=MODEL_DIR,
        local_dir=MODEL_DIR,
        local_dir_use_symlinks=False
    )
else:
    print(f"✅ Model already available at: {MODEL_PATH}")
    model_path = MODEL_PATH

# === Optimal thread usage ===
logical_cores = psutil.cpu_count(logical=True)
physical_cores = psutil.cpu_count(logical=False)
recommended_threads = 4

# === Load the model ===
llm = Llama(
    model_path=model_path,
    n_ctx=2048,  # Can increase depending on memory
    n_threads=recommended_threads,
    use_mlock=True,  # lock model in RAM for faster access
    n_gpu_layers=0,  # CPU only, use >0 if GPU is present
    chat_format="chatml",  # for Hermes 2
    verbose=False
)

@app.get("/")
def root():
    return {"message": "✅ Hermes 2 Mistral API is live and optimized!"}

@app.get("/get_sys")
def get_sys_specs():
    memory = psutil.virtual_memory()
    return {
        "CPU": {
            "physical_cores": physical_cores,
            "logical_cores": logical_cores,
            "max_freq_mhz": psutil.cpu_freq().max,
            "cpu_usage": psutil.cpu_percent(interval=1)
        },
        "RAM": {
            "total_GB": round(memory.total / (1024 ** 3), 2),
            "available_GB": round(memory.available / (1024 ** 3), 2),
            "usage_percent": memory.percent
        },
        "System": {
            "platform": platform.platform(),
            "architecture": platform.machine(),
            "python": platform.python_version()
        }
    }

@app.get("/process_list")
def process_list():
    time.sleep(1)  # Let CPU settle
    processes = []
    for proc in psutil.process_iter(['pid', 'name']):
        try:
            cpu = proc.cpu_percent()
            if cpu > 10:
                processes.append({
                    "pid": proc.pid,
                    "name": proc.name(),
                    "cpu_percent": cpu
                })
        except (psutil.NoSuchProcess, psutil.AccessDenied):
            pass
    return {"heavy_processes": processes}

@app.post("/generate")
async def generate(request: Request):
    data = await request.json()
    prompt = data.get("prompt", "").strip()
    print("🧾 Prompt received:", prompt)

    response = llm.create_chat_completion(
        messages=[
            {"role": "system", "content": "You are a helpful AI assistant."},
            {"role": "user", "content": prompt}
        ],
        max_tokens=2048,
        temperature=0.7,
        stop=["</s>"]
    )

    llm.reset()  # Free memory after response

    return {
        "response": response["choices"][0]["message"]["content"].strip()
    }