app.py

from fastapi import FastAPI, BackgroundTasks
from pydantic import BaseModel
from llama_cpp import Llama
import uuid

app = FastAPI()

# Загружаем Llama-3-8B (сжатую до Q4_K_M — это ~5ГБ)
# Она влезет в 16ГБ ОЗУ и будет на порядок умнее GPT-2 и TinyLlama
llm = Llama.from_pretrained(
    repo_id="MaziyarPanahi/Llama-3-8B-Instruct-v0.1-GGUF",
    filename="*Q4_K_M.gguf",
    n_ctx=2048,  # Контекст (память)
    n_threads=2  # Используем оба ядра твоего бесплатного CPU
)

tasks = {}

class GenerateRequest(BaseModel):
    prompt: str
    system_prompt: str = "Ты — опытный ИИ-помощник. Отвечай четко и только на русском языке."
    custom_id: str = None
    max_tokens: int = 256
    temperature: float = 0.7

def run_ai(task_id, prompt, system_prompt, max_tokens, temp):
    try:
        # Форматируем запрос под стандарт Llama 3
        full_prompt = f"<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>\n\n{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
        
        output = llm(
            full_prompt,
            max_tokens=max_tokens,
            temperature=temp,
            stop=["<|eot_id|>", "<|end_of_text|>"],
            echo=False
        )
        
        tasks[task_id]["result"] = output["choices"][0]["text"].strip()
        tasks[task_id]["status"] = "ready"
    except Exception as e:
        tasks[task_id] = {"status": "error", "result": str(e)}

@app.post("/generate")
async def generate(req: GenerateRequest, background_tasks: BackgroundTasks):
    task_id = req.custom_id if req.custom_id else str(uuid.uuid4())
    tasks[task_id] = {"status": "pending", "result": ""}
    
    background_tasks.add_task(
        run_ai, 
        task_id, 
        req.prompt, 
        req.system_prompt, 
        req.max_tokens, 
        req.temperature
    )
    return {"task_id": task_id}

@app.get("/check/{task_id}")
async def check(task_id: str):
    data = tasks.get(task_id)
    if not data:
        return {"task_id": task_id, "status": "not_found", "result": None}
    return {"task_id": task_id, "status": data["status"], "result": data["result"]}

@app.get("/")
async def root():
    return {"status": "running", "model": "Llama-3-8B-Instruct-GGUF"}