app.py

"""
╔══════════════════════════════════════════════════════════════════╗
║  ATLAS-3 THINKING — Server A                                     ║
║  HuggingFace Space : Vincelil/Atlas3                             ║
║  FastAPI + ORO T3 (1000 unités vectorisées)                      ║
║  Interroge Vincelil/Atlasdata via streaming                      ║
╚══════════════════════════════════════════════════════════════════╝
"""

import asyncio
import time
from contextlib import asynccontextmanager

from fastapi import FastAPI, HTTPException
from fastapi.responses import HTMLResponse, JSONResponse
from fastapi.staticfiles import StaticFiles
from pydantic import BaseModel

from oro_t3 import OroT3Engine
from faiss_index import FaissIndexer

# ── Démarrage : construction de l'index en arrière-plan ───────────
indexer = FaissIndexer()
engine  = OroT3Engine(indexer)


@asynccontextmanager
async def lifespan(app: FastAPI):
    # Lance l'indexation FAISS en tâche de fond au démarrage
    asyncio.create_task(indexer.build_async())
    yield
    indexer.stop()


app = FastAPI(
    title="ATLAS-3 Thinking",
    description="ORO T3 — 1000 unités vectorisées sur FineWeb 15T",
    lifespan=lifespan,
)


# ── Modèles Pydantic ──────────────────────────────────────────────
class QueryRequest(BaseModel):
    query: str
    top_k: int = 5


class QueryResponse(BaseModel):
    query:       str
    output:      str
    oro_results: list
    faiss_hits:  list
    index_size:  int
    tick:        int
    duration_ms: float


# ── Routes ────────────────────────────────────────────────────────

@app.get("/", response_class=HTMLResponse)
async def root():
    with open("templates/index.html", encoding="utf-8") as f:
        return f.read()


@app.post("/query", response_model=QueryResponse)
async def query(req: QueryRequest):
    if not req.query.strip():
        raise HTTPException(400, "Query vide")
    t0 = time.perf_counter()
    result = await engine.process(req.query, req.top_k)
    result["duration_ms"] = round((time.perf_counter() - t0) * 1000, 2)
    return JSONResponse(result)


@app.get("/status")
async def status():
    return {
        "index_size":   indexer.size(),
        "index_ready":  indexer.is_ready(),
        "index_pct":    indexer.progress_pct(),
        "oro_tick":     engine.tick,
        "oro_count":    engine.n_oro,
        "dataset":      "Vincelil/Atlasdata",
    }


@app.get("/oro/apprentissage")
async def apprentissage():
    return engine.get_learning_state()