Add /api/ai/probe diagnostic to find fast available chat model

ai_ext.py

@@ -528,6 +528,30 @@ def pollinations_image_url(topic: str) -> str:
     return "https://image.pollinations.ai/prompt/" + quote(prompt, safe="") + "?width=1024&height=576&nologo=true"
 
 
+@app.get("/api/ai/probe")
+async def api_ai_probe():
+    """Diagnostic: test which chat models actually work on this token + their latency."""
+    import time as _t
+    tok = _hf_token()
+    out = []
+    cand = QWEN_TEXT_MODELS + [QWEN_VL_MODEL, "Qwen/Qwen2.5-VL-7B-Instruct",
+                               "meta-llama/Llama-3.1-8B-Instruct", "google/gemma-2-9b-it",
+                               "mistralai/Mistral-7B-Instruct-v0.3", "Qwen/Qwen2.5-32B-Instruct"]
+    seen = set()
+    for m in cand:
+        if not m or m in seen:
+            continue
+        seen.add(m)
+        t0 = _t.time()
+        try:
+            c = AsyncInferenceClient(provider="auto", api_key=tok, timeout=40)
+            r = await c.chat_completion(model=m, messages=[{"role": "user", "content": "Trả lời đúng 1 từ: xin chào"}], max_tokens=10)
+            out.append({"model": m, "ok": True, "sec": round(_t.time() - t0, 1), "txt": (r.choices[0].message.content or "")[:30]})
+        except Exception as e:
+            out.append({"model": m, "ok": False, "sec": round(_t.time() - t0, 1), "err": (type(e).__name__ + ": " + str(e))[:140]})
+    return JSONResponse({"results": out})
+
+
 # ===== QWEN AI (strict, concise) =====
 async def qwen_generate(prompt: str, image_url: Optional[str] = None, max_tokens: int = 500, image_urls: Optional[List[str]] = None):
     global LAST_QWEN_ERROR, HF_TOKEN