Upload 2 files

app.py

@@ -1,23 +1,23 @@
 """
 Bassma v2 inference endpoint — FastAPI on HuggingFace Spaces.
 
+Pure model server. The Telegram bot lives on Vercel; that side calls
+this /verify endpoint with a shared API key.
+
 Endpoints:
-  GET  /             — health check (also used by keep-alive cron)
-  POST /verify       — JSON API for the Next.js frontend
-  POST /telegram     — Telegram webhook for the bot
+  GET  /        — health check (also used by keep-alive cron)
+  POST /verify  — JSON API for the Next.js frontend AND Vercel bot route
 
 Env (set as Space Secrets in the HF UI):
-  HF_TOKEN             — read token to pull the private model at startup
-  BASSMA_API_KEY       — shared secret with the Next.js app (X-API-KEY)
-  TELEGRAM_BOT_TOKEN   — bot token from @BotFather (NEW)
+  HF_TOKEN        — read token to pull the private model at startup
+  BASSMA_API_KEY  — shared secret with anything that calls /verify
+                    (frontend, Vercel /api/telegram, etc.)
 """
 
 import os
-import time
 
 import numpy as np
-import requests
-from fastapi import FastAPI, Header, HTTPException, Request
+from fastapi import FastAPI, Header, HTTPException
 from pydantic import BaseModel
 from sentence_transformers import SentenceTransformer
 
@@ -25,7 +25,6 @@ MODEL_ID = "ziadabdullah/bassma-v1"
 
 HF_TOKEN = os.environ.get("HF_TOKEN")
 API_KEY = os.environ.get("BASSMA_API_KEY")
-TG_TOKEN = os.environ.get("TELEGRAM_BOT_TOKEN")
 
 print("Loading model...", flush=True)
 model = SentenceTransformer(MODEL_ID, token=HF_TOKEN)
@@ -39,14 +38,6 @@ def normalize(text: str) -> str:
     )
 
 
-def cosine(a: str, b: str) -> float:
-    embs = model.encode(
-        [normalize(a), normalize(b)],
-        normalize_embeddings=True, convert_to_numpy=True,
-    )
-    return float(np.dot(embs[0], embs[1]))
-
-
 class VerifyRequest(BaseModel):
     text_a: str
     text_b: str
@@ -70,105 +61,8 @@ def verify(req: VerifyRequest, x_api_key: str = Header(default="")):
         raise HTTPException(status_code=401, detail="invalid api key")
     if len(req.text_a.strip()) < 50 or len(req.text_b.strip()) < 50:
         raise HTTPException(status_code=400, detail="texts too short")
-    return VerifyResponse(similarity=cosine(req.text_a, req.text_b))
-
-
-# ------------------------------------------------------------------
-# Telegram bot — two-message flow
-# ------------------------------------------------------------------
-# Per-chat state lives in process memory. When the Space restarts the
-# state is lost; users just resend their first text. Good enough for a
-# stateless free-tier Space.
-
-TG_STATE: dict[int, dict] = {}
-TG_MIN_CHARS = 280
-
-
-def tg_send(chat_id: int, text: str) -> None:
-    """Reply via Telegram bot API with retry on transient SSL/network errors."""
-    if not TG_TOKEN:
-        return
-    url = f"https://api.telegram.org/bot{TG_TOKEN}/sendMessage"
-    data = {"chat_id": chat_id, "text": text, "parse_mode": "HTML"}
-    for attempt in range(3):
-        try:
-            r = requests.post(url, data=data, timeout=30)
-            if r.status_code == 200:
-                return
-            print(f"[tg] sendMessage HTTP {r.status_code}: {r.text[:200]}", flush=True)
-        except Exception as e:
-            print(f"[tg] sendMessage attempt {attempt+1} failed: {e}", flush=True)
-            time.sleep(1.5)
-
-
-HELP_AR = (
-    "أهلاً 👋 أنا <b>بصمة</b> — أكشف ما إذا كان نصّان عربيّان كتبهما "
-    "نفس الشخص.\n\n"
-    "أرسل النصّ الأول الآن (٢٨٠ حرفاً على الأقل).\n"
-    "ثم سأطلب منك النصّ الثاني، وسأعطيك نسبة التطابق.\n\n"
-    "/reset — لمسح ما أرسلته والبدء من جديد"
-)
-
-
-def verdict_ar(sim: float) -> str:
-    if sim > 0.85:
-        return "✅ <b>مؤشّر قوي على نفس الكاتب</b>"
-    if sim > 0.70:
-        return "🟡 تقارب أسلوبي ملحوظ — الأدلّة غير حاسمة"
-    if sim > 0.50:
-        return "🟠 تباين أسلوبي مع نقاط تشابه"
-    return "❌ <b>أسلوبان مختلفان — مؤشّر قوي على اختلاف الكاتب</b>"
-
-
-@app.post("/telegram")
-async def telegram_webhook(req: Request):
-    update = await req.json()
-    msg = update.get("message") or update.get("edited_message")
-    if not msg:
-        return {"ok": True}
-
-    chat_id = msg["chat"]["id"]
-    text = (msg.get("text") or "").strip()
-
-    if text in ("/start", "/help"):
-        TG_STATE.pop(chat_id, None)
-        tg_send(chat_id, HELP_AR)
-        return {"ok": True}
-
-    if text == "/reset":
-        TG_STATE.pop(chat_id, None)
-        tg_send(chat_id, "تم المسح. أرسل النصّ الأول.")
-        return {"ok": True}
-
-    if len(text) < TG_MIN_CHARS:
-        tg_send(
-            chat_id,
-            f"النصّ قصير ({len(text)} حرف). أحتاج {TG_MIN_CHARS} حرفاً على الأقل.",
-        )
-        return {"ok": True}
-
-    state = TG_STATE.get(chat_id)
-    if not state or not state.get("first"):
-        TG_STATE[chat_id] = {"first": text}
-        tg_send(chat_id, f"حفظت النصّ الأول ({len(text)} حرف). أرسل النصّ الثاني.")
-        return {"ok": True}
-
-    first = state["first"]
-    TG_STATE.pop(chat_id, None)
-
-    tg_send(chat_id, "جارٍ التحليل…")
-    try:
-        sim = cosine(first, text)
-    except Exception as e:
-        print(f"[tg] verify error: {e}", flush=True)
-        tg_send(chat_id, "تعذّر التحليل. حاول مجدّداً.")
-        return {"ok": True}
-
-    pct = max(0.0, min(1.0, sim)) * 100
-    reply = (
-        f"<b>نسبة التطابق:</b> {pct:.1f}%\n"
-        f"{verdict_ar(sim)}\n\n"
-        "<i>أداة مساعدة وليست دليلاً قانونياً.</i>"
+    embs = model.encode(
+        [normalize(req.text_a), normalize(req.text_b)],
+        normalize_embeddings=True, convert_to_numpy=True,
     )
-    tg_send(chat_id, reply)
-    return {"ok": True}
+    return VerifyResponse(similarity=float(np.dot(embs[0], embs[1])))

requirements.txt

@@ -2,4 +2,3 @@ fastapi==0.115.0
 uvicorn[standard]==0.30.6
 sentence-transformers==3.4.1
 numpy<2.0
-requests==2.32.3