Update app.py

app.py

@@ -13,7 +13,7 @@ logging.basicConfig(level=logging.INFO)
 # Loads from HF Hub at runtime — no local storage needed in the Space
 MERGED_MODEL_DIR = os.environ.get("MODEL_DIR", "SimpleCodeAI/glm-ocr-finetuned")
 
-RENDER_SCALE = 3.0
+
 PAD_LEFT_FRAC = 0.035
 PAD_RIGHT_FRAC = 0.10
 PAD_TOP_FRAC = 0.018
@@ -27,8 +27,9 @@ UNSHARP_PERCENT = 76
 UNSHARP_THRESHOLD = 1
 
 PAGE_PNG_COMPRESS_LEVEL = 3
-MAX_NEW_TOKENS = 2048
-MAX_IMAGE_SIDE = 1344  # resize longest side to this before inference
+RENDER_SCALE = 2.0        # was 3.0 — reduce render size so resize isn't as aggressive
+MAX_IMAGE_SIDE = 1568     # was 1344 — allow slightly larger input to model
+MAX_NEW_TOKENS = 3000  # resize longest side to this before inference
 
 # ── Model singleton ───────────────────────────────────────────────────────────
 _model = None