Pass Qwen prompt through async job

app.py

@@ -517,10 +517,11 @@ def _run_benchmark_job(job_id, kind, model_filter=None, force=False, request=Non
             payload = {"status": "complete", "result": result}
         elif kind == "qwen38":
             request = request or _BENCH_JOBS.get(job_id, {}).get("request", {})
+            qwen_prompt = request.get("prompt") or model_filter or ""
             with _BENCH_LOCK:
                 result = _generate_qwen38(
-                    model_filter,
-                    request.get("max_new_tokens", 64),
+                    qwen_prompt,
+                    request.get("max_new_tokens", 256),
                     request.get("temperature", 0.7),
                     request.get("thinking", False),
                 )
@@ -601,7 +602,7 @@ def _generate_qwen38(prompt, max_new_tokens=256, temperature=0.7, thinking=True)
     processor, model = _load_qwen38()
     max_new_tokens = max(1, min(256, int(max_new_tokens)))
     temperature = max(0.0, min(2.0, float(temperature)))
-    content = prompt
+    content = str(prompt).strip()
     messages = [{"role": "user", "content": [{"type": "text", "text": content}]}]
     try:
         text = processor.apply_chat_template(
@@ -786,6 +787,7 @@ def _patched_create_app(blocks, **kwargs):
             "qwen38",
             None,
             request={
+                "prompt": prompt,
                 "max_new_tokens": payload.get("max_new_tokens", 256),
                 "temperature": payload.get("temperature", 0.7),
                 "thinking": payload.get("thinking", True),