Fix short-input error: check token length in main process before @spaces.GPU call

app.py

@@ -45,7 +45,7 @@ MEMCLR, NONCLR = "#f58518", "#4c78a8"      # memorized (orange) vs not-memorized
 _ON_ZEROGPU = bool(os.environ.get("SPACES_ZERO_GPU"))
 DEVICE = "cuda" if (_ON_ZEROGPU or torch.cuda.is_available()) else "cpu"
 try:
-    from byo_features import BYOExtractor
+    from byo_features import BYOExtractor, WINDOW_LEN
     EXTRACTOR = BYOExtractor(BASE_MODEL, device=DEVICE)
     _LOAD_ERR = None
 except Exception as e:
@@ -136,6 +136,12 @@ def predict(text, clf_name):
         return "Please enter some input text first.", None
     if EXTRACTOR is None:
         return f"⚠️ Base model unavailable: {_LOAD_ERR}", None
+    # length check in the main process (CPU tokenization); a ValueError raised inside the
+    # @spaces.GPU worker would not survive the process boundary as a catchable ValueError.
+    n_tokens = len(EXTRACTOR.tok(text, truncation=True, max_length=WINDOW_LEN)["input_ids"])
+    if n_tokens < WINDOW_LEN:
+        return (f"⚠️ The input should be at least {WINDOW_LEN} tokens (or about 75 words); "
+                f"you entered {n_tokens} tokens."), None
     try:
         feats = _compute_features(text)
     except ValueError as e: