Upload app.py with huggingface_hub

app.py

@@ -267,18 +267,12 @@ def solve(
             pos = full_text.rfind(prompt)
             completion = full_text[pos + len(prompt):] if pos != -1 else full_text
         completion = strip_after_stop_text(completion, STOP_STRINGS)
-        return _clean(completion)
+        return completion.strip()
 
-    # Raw mode: return the full continued expression.
+    # Raw mode: return the full continued expression, matching the reference
+    # inference script's behavior exactly (strip only at [IM_END]/[IM_START]).
     completion = strip_after_stop_text(full_text, STOP_STRINGS)
-    return _clean(completion)
-
-
-def _clean(text: str) -> str:
-    """Display-only cleanup: drop the training-artifact `/no think` control
-    marker the model sometimes appends. Does not affect generation."""
-    text = strip_after_stop_text(text, [NO_THINK])
-    return text.strip()
+    return completion.strip()
 
 
 CSS = """