21

app.py

@@ -1,7 +1,7 @@
 # app.py
 # PapuaTranslate — Flask + SQLAlchemy (Supabase/SQLite) + mT5/LoRA (lazy)
 # Fitur: ProxyFix, cookie Spaces, session permanent, /diag, preload, fallback,
-#        strip BOM JSON, sanitize adapter_config untuk kompatibilitas PEFT lama.
+#        strip BOM JSON, sanitize adapter_config, endpoint /history + homepage responsive.
 
 import os, re, json, codecs, pathlib, logging, threading, traceback, inspect
 from datetime import datetime, timezone, timedelta
@@ -10,13 +10,16 @@ from flask import Flask, render_template, request, redirect, url_for, session, j
 from werkzeug.middleware.proxy_fix import ProxyFix
 from huggingface_hub import snapshot_download
 
-# ========== Logging ==========
+# ---------- Logging ----------
 logging.basicConfig(level=logging.INFO, format="%(asctime)s | %(levelname)s | %(message)s")
 log = logging.getLogger("papua-app")
 
-# ========== Flask ==========
+# ---------- Flask ----------
 app = Flask(__name__, template_folder="frontend", static_folder="static")
+# trust HF proxy (HTTPS/host), penting untuk cookie Secure & redirect
 app.wsgi_app = ProxyFix(app.wsgi_app, x_for=1, x_proto=1, x_host=1)
+
+# session config (HF Spaces iframe-friendly)
 app.config.update(
     SECRET_KEY=os.getenv("SECRET_KEY", "dev-secret-change-me"),
     SESSION_COOKIE_NAME="hfspace_session",
@@ -28,11 +31,11 @@ app.config.update(
 )
 app.permanent_session_lifetime = timedelta(hours=8)
 
-# ========== Feature Flags via ENV ==========
+# ---------- 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")
 
-# ========== Database (SQLAlchemy) ==========
+# ---------- Database ----------
 from sqlalchemy import create_engine, Column, Integer, Text, DateTime, ForeignKey, func
 from sqlalchemy.orm import declarative_base, sessionmaker, scoped_session, relationship
 
@@ -75,7 +78,7 @@ try:
 except Exception as e:
     log.exception("[DB] init error: %s", e)
 
-# ========== Auth Helpers ==========
+# ---------- Auth helpers ----------
 from werkzeug.security import generate_password_hash, check_password_hash
 def set_password(user: User, raw: str): user.pass_hash = generate_password_hash(raw)
 def verify_password(user: User, raw: str) -> bool:
@@ -90,7 +93,7 @@ def login_required(fn):
         return fn(*args, **kwargs)
     return _wrap
 
-# ========== Prenorm ==========
+# ---------- Prenorm ----------
 PAPUA_MAP = {
     r"\bsa\b": "saya", r"\bko\b": "kamu", r"\btra\b": "tidak", r"\bndak\b": "tidak",
     r"\bmo\b": "mau", r"\bpu\b": "punya", r"\bsu\b": "sudah", r"\bkong\b": "kemudian",
@@ -101,7 +104,7 @@ def prenorm(text: str) -> str:
     for pat, repl in PAPUA_MAP.items(): t = re.sub(pat, repl, t, flags=re.IGNORECASE)
     return t
 
-# ========== Model (lazy) + Strip BOM + Sanitize adapter_config ==========
+# ---------- Model (lazy) + strip BOM + sanitize adapter ----------
 BASE_MODEL_ID = os.getenv("BASE_MODEL_ID", "google/mt5-small")
 ADAPTER_ID    = os.getenv("ADAPTER_ID", "")
 DEVICE        = "cuda" if os.getenv("DEVICE", "cpu") == "cuda" else "cpu"
@@ -125,17 +128,11 @@ def _strip_bom_in_dir(root_dir: str):
             log.warning(f"[BOM] skip {p}: {e}")
 
 def _sanitize_adapter_config(adapter_dir: str):
-    """
-    Buang kunci tidak dikenal oleh peft.LoraConfig saat ini.
-    Menangani error seperti: LoraConfig.__init__() got an unexpected keyword argument 'e_rank_pattern'
-    """
     try:
-        from peft import LoraConfig  # versi yang sedang terpasang di runtime
+        from peft import LoraConfig
     except Exception as e:
         log.warning(f"[SAN] Tidak bisa import LoraConfig: {e}")
         return
-
-    sig = None
     try:
         sig = inspect.signature(LoraConfig.__init__)
         allowed = set(p.name for p in sig.parameters.values())
@@ -145,12 +142,10 @@ def _sanitize_adapter_config(adapter_dir: str):
 
     cfg_path = pathlib.Path(adapter_dir) / "adapter_config.json"
     if not cfg_path.exists():
-        # beberapa repo pakai nama lain; normalnya adapter_config.json
-        for alt in ("adapter_config.json", "adapter_config_0.json", "config.json"):
-            candidate = pathlib.Path(adapter_dir) / alt
-            if candidate.exists():
-                cfg_path = candidate
-                break
+        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
@@ -162,7 +157,6 @@ def _sanitize_adapter_config(adapter_dir: str):
         log.warning(f"[SAN] gagal baca adapter_config: {e}")
         return
 
-    # filter hanya key yang diizinkan oleh LoraConfig saat ini
     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:
@@ -171,7 +165,7 @@ def _sanitize_adapter_config(adapter_dir: str):
             json.dump(cleaned, f, ensure_ascii=False, indent=2)
 
 def _load_model():
-    """Download ke /tmp, strip BOM, sanitize adapter_config, lalu load dari path lokal."""
+    """Download ke /tmp, strip BOM, sanitize adapter_config, lalu load."""
     global TOK, MODEL, _MODEL_READY, _MODEL_ERROR
     try:
         log.info("[MODEL] downloading base=%s adapter=%s", BASE_MODEL_ID, ADAPTER_ID or "-")
@@ -199,19 +193,12 @@ def _load_model():
         from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
         from peft import PeftModel
 
-        log.info("[MODEL] loading tokenizer from %s", base_dir)
         TOK = AutoTokenizer.from_pretrained(base_dir)
-
-        log.info("[MODEL] loading base model from %s", base_dir)
         base = AutoModelForSeq2SeqLM.from_pretrained(base_dir)
 
-        if adapter_dir:
-            log.info("[MODEL] attaching adapter from %s", adapter_dir)
-            MODEL = PeftModel.from_pretrained(base, adapter_dir)
-        else:
-            MODEL = base
-
+        MODEL = PeftModel.from_pretrained(base, adapter_dir) if adapter_dir else base
         MODEL.eval().to(DEVICE)
+
         _MODEL_READY = True
         _MODEL_ERROR = None
         log.info("[MODEL] ready on %s", DEVICE)
@@ -221,7 +208,7 @@ def _load_model():
         log.exception("[MODEL] load error")
 
 def get_model():
-    global MODEL, _MODEL_READY
+    global MODEL
     if MODEL is None:
         with _MODEL_LOCK:
             if MODEL is None:
@@ -246,18 +233,16 @@ def translate_with_model(text: str, max_new_tokens: int = 48) -> str:
     return tok.decode(out[0], skip_special_tokens=True)
 
 def _preload_thread():
-    try:
-        _load_model()
-    except Exception:
-        pass
+    try: _load_model()
+    except Exception: pass
 
 if PRELOAD_MODEL:
     threading.Thread(target=_preload_thread, daemon=True).start()
 
-# ========== Utils / Logging ==========
+# ---------- Utils / Logging ----------
 @app.before_request
 def _log_req():
-    if request.path not in ("/health", "/ping", "/favicon.ico"):
+    if request.path not in ("/health","/ping","/favicon.ico"):
         log.info("[REQ] %s %s", request.method, request.path)
 
 @app.errorhandler(Exception)
@@ -265,7 +250,7 @@ def _err(e):
     log.exception("Unhandled error")
     return "Internal Server Error", 500
 
-# ========== Debug & Diag ==========
+# ---------- Debug & Diag ----------
 @app.get("/diag")
 def diag():
     import sys
@@ -276,7 +261,6 @@ def diag():
         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(),
@@ -300,7 +284,7 @@ def dbg_set():
 def dbg_get():
     return {"uid": session.get("uid"), "email": session.get("email")}
 
-# ========== Routes ==========
+# ---------- Auth & Pages ----------
 @app.get("/health")
 @app.get("/ping")
 def health():
@@ -357,24 +341,36 @@ def index():
                  .filter(Translation.user_id == uid)
                  .order_by(Translation.id.desc())
                  .limit(10).all())
-        data = [{"src": it.src, "mt": it.mt, "created_at": it.created_at} for it in items]
-    return render_template("index.html", user=session.get("email"), data=data, device=DEVICE)
+        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")
 
+# ---------- API ----------
+@app.get("/history")
+def api_history():
+    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())
+        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
-
     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
-
     try:
         clean = prenorm(text)
         mt = f"[FAKE] {clean}" if FALLBACK_TRANSLATE else translate_with_model(clean, max_new_tokens=max_new)
@@ -387,5 +383,6 @@ def api_translate():
         log.error(traceback.format_exc())
         return jsonify({"ok": False, "error": f"{type(e).__name__}: {e}"}), 500
 
+# ---------- Run ----------
 if __name__ == "__main__":
     app.run(host="0.0.0.0", port=int(os.getenv("PORT", "7860")), debug=False)