feat: use larger model

app.py

@@ -11,8 +11,8 @@ from transformers import (
 # ▸ For a causal code model (no T5 errors):
 # MODEL_ID = "Salesforce/codegen-350M-multi"
 # ▸ Or, for a seq‑to‑seq model:
-# MODEL_ID = "google/flan-t5-base"
-MODEL_ID = "google/flan-t5-small"
+MODEL_ID = "google/flan-t5-base"
+# MODEL_ID = "google/flan-t5-small"
 # MODEL_ID = "Salesforce/codegen-350M-multi"
 
 # ── 2) Load tokenizer + model ────────────────────────────────────────
@@ -46,10 +46,17 @@ def review_code(diff: str, guidelines: str):
         f"DIFF:\n{diff}\n\n"
         f"GUIDELINES:\n{guidelines}\n\n"
         "OUTPUT FORMAT EXAMPLE:\n"
-        '[{"line":12,"comment":"…"}]\n'
+        '[{"line":12,"comment":"…"}]\n\n'
+        "<<<END_JSON>>>\n"
     )
 
+    # Run the model
     out = pipe(prompt)[0]["generated_text"]
+
+    # Truncate at our stop marker, if present
+    if "<<<END_JSON>>>" in out:
+        out = out.split("<<<END_JSON>>>")[0]
+
     start = out.find("[")
     end   = out.rfind("]") + 1
     if start < 0 or end < 0: