Upload app.py with huggingface_hub

app.py

@@ -643,18 +643,41 @@ with gr.Blocks(title="needle3 logits") as demo:
         cmp_btn.click(compare_backends, None, cmp_out, api_name="compare")
         tun_btn = gr.Button("tuned adapter only (single state)")
         tun_out = gr.JSON(label="tuned response")
-        tun_btn.click(
-            lambda s: answer_tuned(s, {
-                "department": {"type": "choice", "instructions": "Which team?",
-                               "criteria": {k: {"description": v, "triggers": TRIG[k]}
-                                            for k, v in
-                                            {"billing": "Payment, charges, invoices, refunds.",
-                                             "technical": "Bugs, errors, outages.",
-                                             "account": "Login, password, access."}.items()}},
-                "refund_requested": {"type": "noul",
-                                     "instructions": "Is the writer asking for a refund?"},
-                "_request_text": s}),
-            [st], tun_out, api_name="tuned")
+        tun_q = gr.Code(label="questions (JSON) -- passed through verbatim", language="json",
+                        value=json.dumps({
+                            "department": {"type": "choice",
+                                           "instructions": "Which team?",
+                                           "criteria": {k: {"description": v, "triggers": TRIG[k]}
+                                                        for k, v in
+                                                        {"billing": "Payment, charges, refunds.",
+                                                         "technical": "Bugs, errors, outages.",
+                                                         "account": "Login, password, access."}.items()}},
+                            "refund_requested": {"type": "noul",
+                                                 "instructions": "Is the writer asking for a refund?"}},
+                            indent=2))
+
+        def _run_tuned(state, questions_text):
+            """Run the tuned adapter on the CALLER'S questions.
+
+            This used to take only the state and hardcode its own two questions. That made
+            the endpoint useless for comparison work -- a head-to-head that sent its own
+            questions was silently answered on a different, fixed set, and the mismatch
+            looked like a model result. Questions are now passed through.
+            """
+            if isinstance(questions_text, str):
+                try:
+                    qs = json.loads(questions_text)
+                except Exception as e:
+                    return {"error": f"questions is not valid JSON: {e}"}
+            else:
+                qs = questions_text
+            if not isinstance(qs, dict) or not qs:
+                return {"error": "questions must be a non-empty JSON object"}
+            qs = dict(qs)
+            qs["_request_text"] = state
+            return answer_tuned(state, qs)
+
+        tun_btn.click(_run_tuned, [st, tun_q], tun_out, api_name="tuned")
 
     with gr.Tab("Method comparison"):
         gr.Markdown("Which scoring method actually separates the three clear cases?")