Updatefe3

app.py

@@ -32,6 +32,12 @@ app.permanent_session_lifetime = timedelta(hours=8)
 # ---------- Feature Flags ----------
 PRELOAD_MODEL      = os.getenv("PRELOAD_MODEL", "true").lower() in ("1","true","yes")
 FALLBACK_TRANSLATE = os.getenv("FALLBACK_TRANSLATE", "false").lower() in ("1","true","yes")
+PUBLIC_APP_URL     = os.getenv("PUBLIC_APP_URL", "").strip()   # ← URL fullscreen (hf.space)
+
+# bikin variabel template global
+@app.context_processor
+def inject_globals():
+    return {"public_app_url": PUBLIC_APP_URL}
 
 # ---------- Database ----------
 from sqlalchemy import create_engine, Column, Integer, Text, DateTime, ForeignKey, func
@@ -93,7 +99,7 @@ def login_required(fn):
         return fn(*args, **kwargs)
     return _wrap
 
-# ---------- Prenorm ----------
+# ---------- Prenorm (Papua) ----------
 PRENORM_LEVEL = os.getenv("PRENORM_LEVEL", "basic").lower()
 PRENORM_DEBUG = os.getenv("PRENORM_DEBUG", "0") == "1"
 
@@ -125,57 +131,37 @@ PRON_MAP = {
     "kitong": "kita", "kitorang": "kita", "kita": "kita", "torang": "kita",
 }
 
-def _normalize_unicode(text: str) -> str:
-    return unicodedata.normalize("NFKC", text)
+def _normalize_unicode(text: str) -> str: return unicodedata.normalize("NFKC", text)
 def _strip_emoji_and_noise(text: str) -> str:
-    text = PUNC_RE.sub(" ", text)
-    text = MULTI_PUNC.sub(r"\1", text)
-    text = DASH_SPACES.sub(r" \1 ", text)
-    return text
-def _normalize_ws(text: str) -> str:
-    return WS_RE.sub(" ", text).strip()
+    text = PUNC_RE.sub(" ", text); text = MULTI_PUNC.sub(r"\1", text)
+    return DASH_SPACES.sub(r" \1 ", text)
+def _normalize_ws(text: str) -> str: return WS_RE.sub(" ", text).strip()
 def _reduce_elongation(token: str) -> str:
     base = token.lower()
     if base in WHITELIST_KEEP_ELONG: return token
     return ELONG_RE.sub(r"\1\1", token)
-def _apply_papua_map(token: str) -> str:
-    low = token.lower()
-    return PAPUA_MAP.get(low, token)
+def _apply_papua_map(token: str) -> str: return PAPUA_MAP.get(token.lower(), token)
 def _handle_pu_constructs(text: str) -> str:
     def repl(m):
-        pron = m.group(1).lower()
-        rest = m.group(2).strip()
-        pron_std = PRON_MAP.get(pron, pron)
-        return f"punya {pron_std} {rest}"
-    return re.sub(r"\b(sa|saya|ko|kamu|dia|dong|kam|kalian|kitong|kitorang|kita|torang)\s*pu\s+([^.,;:!?]+)",
-                  repl, text, flags=re.IGNORECASE)
+        pron = m.group(1).lower(); rest = m.group(2).strip()
+        pron_std = PRON_MAP.get(pron, pron); return f"punya {pron_std} {rest}"
+    return re.sub(r"\b(sa|saya|ko|kamu|dia|dong|kam|kalian|kitong|kitorang|kita|torang)\s*pu\s+([^.,;:!?]+)", repl, text, flags=re.IGNORECASE)
 def _token_level_ops(text: str, aggressive: bool) -> str:
-    tokens = text.split()
-    out = []
-    for t in tokens:
+    toks = text.split(); out = []
+    for t in toks:
         t2 = _reduce_elongation(t) if aggressive else t
-        t3 = _apply_papua_map(t2)
-        out.append(t3)
+        out.append(_apply_papua_map(t2))
     return " ".join(out)
 
 def papua_prenorm(inp: str, level: str = "basic", return_trace: bool = False):
-    trace = {"level": level}
-    if level == "off":
-        return (inp, trace) if return_trace else inp
-    s0 = inp
-    s1 = _normalize_unicode(s0)
+    if level == "off": return (inp, {}) if return_trace else inp
+    s1 = _normalize_unicode(inp)
     s2 = _strip_emoji_and_noise(s1) if level == "aggressive" else s1
     s3 = _normalize_ws(s2)
     s4 = _handle_pu_constructs(s3)
     s5 = _token_level_ops(s4, aggressive=(level == "aggressive"))
     s6 = _normalize_ws(s5)
-    if return_trace:
-        trace.update({"original": s0, "unicode_norm": s1,
-                      "strip_noise": s2 if level == "aggressive" else "(skip)",
-                      "ws_norm_1": s3, "pu_constructs": s4,
-                      "token_ops": s5, "final": s6})
-        return s6, trace
-    return s6
+    return (s6, {"final": s6}) if return_trace else s6
 
 def prenorm(text: str) -> str:
     if PRENORM_DEBUG:
@@ -189,20 +175,15 @@ BASE_MODEL_ID = os.getenv("BASE_MODEL_ID", "amosnbn/cendol-mt5-base-inst")
 ADAPTER_ID    = os.getenv("ADAPTER_ID", "amosnbn/papua-lora-ckpt-168")
 DEVICE        = "cuda" if os.getenv("DEVICE", "cpu") == "cuda" else "cpu"
 
-TOK = None
-MODEL = None
-_MODEL_LOCK = threading.Lock()
-_MODEL_READY = False
-_MODEL_ERROR = None
+TOK = None; MODEL = None
+_MODEL_LOCK = threading.Lock(); _MODEL_READY = False; _MODEL_ERROR = None
 
 def _strip_bom_in_dir(root_dir: str):
     root = pathlib.Path(root_dir)
     for p in root.rglob("*.json"):
         try:
-            with codecs.open(p, "r", encoding="utf-8-sig") as f:
-                data = json.load(f)
-            with open(p, "w", encoding="utf-8") as f:
-                json.dump(data, f, ensure_ascii=False, indent=2)
+            with codecs.open(p, "r", encoding="utf-8-sig") as f: data = json.load(f)
+            with open(p, "w", encoding="utf-8") as f: json.dump(data, f, ensure_ascii=False, indent=2)
             log.info(f"[BOM] stripped: {p}")
         except Exception as e:
             log.warning(f"[BOM] skip {p}: {e}")
@@ -210,39 +191,19 @@ def _strip_bom_in_dir(root_dir: str):
 def _sanitize_adapter_config(adapter_dir: str):
     try:
         from peft import LoraConfig
+        sig = inspect.signature(LoraConfig.__init__); allowed = set(p.name for p in sig.parameters.values())
+        cfg_path = pathlib.Path(adapter_dir) / "adapter_config.json"
+        if not cfg_path.exists():
+            for alt in ("adapter_config.json","adapter_config_0.json","config.json"):
+                c = pathlib.Path(adapter_dir)/alt
+                if c.exists(): cfg_path = c; break
+        if not cfg_path.exists(): return
+        with codecs.open(cfg_path, "r", encoding="utf-8-sig") as f: cfg = json.load(f)
+        cleaned = {k:v for k,v in cfg.items() if k in allowed}
+        if set(cleaned.keys()) != set(cfg.keys()):
+            with open(cfg_path, "w", encoding="utf-8") as f: json.dump(cleaned, f, ensure_ascii=False, indent=2)
     except Exception as e:
-        log.warning(f"[SAN] Tidak bisa import LoraConfig: {e}")
-        return
-    try:
-        sig = inspect.signature(LoraConfig.__init__)
-        allowed = set(p.name for p in sig.parameters.values())
-    except Exception as e:
-        log.warning(f"[SAN] Tidak bisa baca signature LoraConfig: {e}")
-        return
-
-    cfg_path = pathlib.Path(adapter_dir) / "adapter_config.json"
-    if not cfg_path.exists():
-        for alt in ("adapter_config.json","adapter_config_0.json","config.json"):
-            c = pathlib.Path(adapter_dir)/alt
-            if c.exists():
-                cfg_path = c; break
-    if not cfg_path.exists():
-        log.warning(f"[SAN] adapter_config tidak ditemukan di {adapter_dir}")
-        return
-
-    try:
-        with codecs.open(cfg_path, "r", encoding="utf-8-sig") as f:
-            cfg = json.load(f)
-    except Exception as e:
-        log.warning(f"[SAN] gagal baca adapter_config: {e}")
-        return
-
-    cleaned = {k: v for k, v in cfg.items() if k in allowed}
-    dropped = [k for k in cfg.keys() if k not in allowed]
-    if dropped:
-        log.info(f"[SAN] drop fields tidak dikenal: {dropped}")
-        with open(cfg_path, "w", encoding="utf-8") as f:
-            json.dump(cleaned, f, ensure_ascii=False, indent=2)
+        log.warning(f"[SAN] skip: {e}")
 
 def _load_model():
     global TOK, MODEL, _MODEL_READY, _MODEL_ERROR
@@ -250,12 +211,10 @@ def _load_model():
         log.info("[MODEL] downloading base=%s adapter=%s", BASE_MODEL_ID, ADAPTER_ID or "-")
         base_dir = snapshot_download(repo_id=BASE_MODEL_ID, local_dir="/tmp/hf_base", local_dir_use_symlinks=False)
         _strip_bom_in_dir(base_dir)
-
         adapter_dir = None
         if ADAPTER_ID:
             adapter_dir = snapshot_download(repo_id=ADAPTER_ID, local_dir="/tmp/hf_adapter", local_dir_use_symlinks=False)
-            _strip_bom_in_dir(adapter_dir)
-            _sanitize_adapter_config(adapter_dir)
+            _strip_bom_in_dir(adapter_dir); _sanitize_adapter_config(adapter_dir)
 
         import torch
         from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
@@ -266,20 +225,17 @@ def _load_model():
         MODEL = PeftModel.from_pretrained(base, adapter_dir) if adapter_dir else base
         MODEL.eval().to(DEVICE)
 
-        _MODEL_READY = True
-        _MODEL_ERROR = None
+        _MODEL_READY = True; _MODEL_ERROR = None
         log.info("[MODEL] ready on %s", DEVICE)
     except Exception as e:
-        _MODEL_READY = False
-        _MODEL_ERROR = f"{type(e).__name__}: {e}"
+        _MODEL_READY = False; _MODEL_ERROR = f"{type(e).__name__}: {e}"
         log.exception("[MODEL] load error")
 
 def get_model():
     global MODEL
     if MODEL is None:
         with _MODEL_LOCK:
-            if MODEL is None:
-                _load_model()
+            if MODEL is None: _load_model()
     return TOK, MODEL
 
 def translate_with_model(text: str, max_new_tokens: int = 48) -> str:
@@ -289,14 +245,8 @@ def translate_with_model(text: str, max_new_tokens: int = 48) -> str:
         raise RuntimeError(f"Model not ready: {_MODEL_ERROR or 'unknown error'}")
     enc = tok([text], return_tensors="pt", truncation=True, max_length=256)
     enc = {k: v.to(DEVICE) for k, v in enc.items()}
-    out = m.generate(
-        **enc,
-        max_new_tokens=int(max_new_tokens),
-        num_beams=4,
-        length_penalty=0.9,
-        no_repeat_ngram_size=3,
-        early_stopping=True,
-    )
+    out = m.generate(**enc, max_new_tokens=int(max_new_tokens), num_beams=4,
+                     length_penalty=0.9, no_repeat_ngram_size=3, early_stopping=True)
     return tok.decode(out[0], skip_special_tokens=True)
 
 def _preload_thread():
@@ -323,34 +273,25 @@ def diag():
     import sys
     try:
         import torch, transformers, peft
-        torch_v = torch.__version__
-        tf_v    = transformers.__version__
-        peft_v  = peft.__version__
+        torch_v = torch.__version__; tf_v = transformers.__version__; peft_v = peft.__version__
     except Exception as e:
         torch_v = tf_v = peft_v = f"import error: {e}"
     return jsonify({
-        "ok": True,
-        "time": datetime.now(timezone.utc).isoformat(),
-        "device": DEVICE,
-        "base_model": BASE_MODEL_ID,
-        "adapter": ADAPTER_ID or None,
-        "model_ready": _MODEL_READY,
-        "model_error": _MODEL_ERROR,
+        "ok": True, "time": datetime.now(timezone.utc).isoformat(), "device": DEVICE,
+        "base_model": BASE_MODEL_ID, "adapter": ADAPTER_ID or None,
+        "model_ready": _MODEL_READY, "model_error": _MODEL_ERROR,
         "versions": {"python": sys.version, "torch": torch_v, "transformers": tf_v, "peft": peft_v},
-        "preload": PRELOAD_MODEL,
-        "prenorm": {"level": PRENORM_LEVEL, "debug": PRENORM_DEBUG},
+        "preload": PRELOAD_MODEL, "prenorm": {"level": PRENORM_LEVEL, "debug": PRENORM_DEBUG},
+        "public_app_url": PUBLIC_APP_URL or None
     })
 
 # ---------- Auth & Pages ----------
-@app.get("/health")
-@app.get("/ping")
-def health():
-    return jsonify({"ok": True, "time": datetime.now(timezone.utc).isoformat()})
+@app.get("/health"); @app.get("/ping")
+def health(): return jsonify({"ok": True, "time": datetime.now(timezone.utc).isoformat()})
 
 @app.get("/login")
 def login_get():
-    if session.get("uid"):  # sudah login → ke home
-        return redirect(url_for("index"))
+    if session.get("uid"): return redirect(url_for("index"))
     return render_template("login.html")
 
 @app.post("/login")
@@ -358,24 +299,17 @@ def login_post():
     email = (request.form.get("email") or "").strip().lower()
     pwd   = request.form.get("password") or ""
     if not email or not pwd:
-        flash("Isi email dan password", "error")
-        return redirect(url_for("login_get"))
-
+        flash("Isi email dan password", "error"); return redirect(url_for("login_get"))
     with SessionLocal() as s:
         u = s.query(User).filter_by(email=email).first()
         if not u or not verify_password(u, pwd):
-            flash("Email atau password salah", "error")
-            return redirect(url_for("login_get"))
-
-        session.permanent = True
-        session["uid"], session["email"] = u.id, u.email
-
+            flash("Email atau password salah", "error"); return redirect(url_for("login_get"))
+        session.permanent = True; session["uid"], session["email"] = u.id, u.email
     return redirect(url_for("index"))
 
 @app.get("/register")
 def register_get():
-    if session.get("uid"):  # sudah login → ke home
-        return redirect(url_for("index"))
+    if session.get("uid"): return redirect(url_for("index"))
     return render_template("register.html")
 
 @app.post("/register")
@@ -383,24 +317,17 @@ def register_post():
     email = (request.form.get("email") or "").strip().lower()
     pwd   = (request.form.get("password") or "")
     if not email or not pwd:
-        flash("Isi email dan password", "error")
-        return redirect(url_for("register_get"))
-
+        flash("Isi email dan password", "error"); return redirect(url_for("register_get"))
     with SessionLocal() as s:
         if s.query(User).filter_by(email=email).first():
-            flash("Email sudah terdaftar", "error")
-            return redirect(url_for("register_get"))
-        u = User(email=email); set_password(u, pwd)
-        s.add(u); s.commit()
-
-    # BEDA DI SINI: tidak auto-login. Wajib login manual.
+            flash("Email sudah terdaftar", "error"); return redirect(url_for("register_get"))
+        u = User(email=email); set_password(u, pwd); s.add(u); s.commit()
     flash("Registrasi berhasil. Silakan login.", "success")
     return redirect(url_for("login_get"))
 
 @app.get("/logout")
 def logout():
-    session.clear()
-    return redirect(url_for("login_get"))
+    session.clear(); return redirect(url_for("login_get"))
 
 @app.get("/")
 @login_required
@@ -409,48 +336,40 @@ def index():
         uid = session["uid"]
         items = (s.query(Translation)
                  .filter(Translation.user_id == uid)
-                 .order_by(Translation.id.desc())
-                 .limit(10).all())
+                 .order_by(Translation.id.desc()).limit(10).all())
         recent = [{"src": it.src, "mt": it.mt, "created_at": it.created_at.strftime("%Y-%m-%d %H:%M")} for it in items]
     return render_template("index.html", logged_in=True, device=DEVICE, recent=recent)
 
 @app.get("/about")
-def about_page():
-    return render_template("about.html")
+def about_page(): return render_template("about.html")
 
 # ---------- API ----------
 @app.get("/history")
 def api_history():
-    if not session.get("uid"):
-        return jsonify({"items": []})
+    if not session.get("uid"): return jsonify({"items": []})
     with SessionLocal() as s:
         uid = session["uid"]
         items = (s.query(Translation)
                  .filter(Translation.user_id == uid)
-                 .order_by(Translation.id.desc())
-                 .limit(10).all())
+                 .order_by(Translation.id.desc()).limit(10).all())
         out = [{"src": it.src, "mt": it.mt, "created_at": it.created_at.strftime("%Y-%m-%d %H:%M")} for it in items]
     return jsonify({"items": out})
 
 @app.post("/translate")
 def api_translate():
-    if not session.get("uid"):
-        return jsonify({"ok": False, "error": "Unauthorized"}), 401
+    if not session.get("uid"): return jsonify({"ok": False, "error": "Unauthorized"}), 401
     payload = request.get_json(silent=True) or {}
     text = (payload.get("text") or "").strip()
     max_new = int(payload.get("max_new_tokens", 48))
-    if not text:
-        return jsonify({"ok": False, "error": "Empty text"}), 400
+    if not text: return jsonify({"ok": False, "error": "Empty text"}), 400
     try:
         clean = prenorm(text)
         mt = f"[FAKE] {clean}" if FALLBACK_TRANSLATE else translate_with_model(clean, max_new_tokens=max_new)
         with SessionLocal() as s:
-            s.add(Translation(user_id=session["uid"], src=text, mt=mt))
-            s.commit()
+            s.add(Translation(user_id=session["uid"], src=text, mt=mt)); s.commit()
         return jsonify({"ok": True, "mt": mt})
     except Exception as e:
-        log.error("[API] translate error: %s", e)
-        log.error(traceback.format_exc())
+        log.error("[API] translate error: %s", e); log.error(traceback.format_exc())
         return jsonify({"ok": False, "error": f"{type(e).__name__}: {e}"}), 500
 
 # ---------- Run ----------

frontend/login.html

@@ -18,6 +18,8 @@
     button{width:100%;margin-top:14px;padding:12px;background:#000;color:#fff;border:0;border-radius:8px;cursor:pointer}
     .note{margin-top:10px;font-size:14px;color:#666;text-align:center}
     .flash{margin-top:10px;padding:10px;border-radius:6px;font-size:14px;background:#fdecec;color:#a02222;border:1px solid #f5b5b5}
+    .hint{margin-top:14px;text-align:center;font-size:14px;color:#555}
+    .hint a{color:#000;text-decoration:underline}
   </style>
 </head>
 <body>
@@ -47,6 +49,12 @@
         <button type="submit">Login</button>
       </form>
       <p class="note">Belum punya akun? <a href="/register">Daftar</a></p>
+
+      {% if public_app_url %}
+        <div class="hint">
+          Masalah login di HP? <a href="{{ public_app_url }}" target="_top">Buka Full App</a>
+        </div>
+      {% endif %}
     </div>
   </div>
 </body>

frontend/register.html

@@ -18,6 +18,8 @@
     button{width:100%;margin-top:14px;padding:12px;background:#000;color:#fff;border:0;border-radius:8px;cursor:pointer}
     .note{margin-top:10px;font-size:14px;color:#666;text-align:center}
     .flash{margin-top:10px;padding:10px;border-radius:6px;font-size:14px;background:#fdecec;color:#a02222;border:1px solid #f5b5b5}
+    .hint{margin-top:14px;text-align:center;font-size:14px;color:#555}
+    .hint a{color:#000;text-decoration:underline}
   </style>
 </head>
 <body>
@@ -47,6 +49,12 @@
         <button type="submit">Daftar</button>
       </form>
       <p class="note">Sudah punya akun? <a href="/login">Login</a></p>
+
+      {% if public_app_url %}
+        <div class="hint">
+          Login di HP kadang terblokir iframe. Setelah daftar, silakan <a href="{{ public_app_url }}" target="_top">Buka Full App</a> lalu login.
+        </div>
+      {% endif %}
     </div>
   </div>
 </body>