probe: more candidate models + full error

ai_ext.py

@@ -534,9 +534,11 @@ async def api_ai_probe():
     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"]
+    extra = os.getenv("PROBE_MODELS", "").split(",")
+    cand = [m.strip() for m in extra if m.strip()] + [
+        "Qwen/Qwen2.5-VL-7B-Instruct", "Qwen/Qwen2.5-VL-3B-Instruct", "Qwen/Qwen2-VL-7B-Instruct",
+        "Qwen/Qwen2.5-VL-72B-Instruct", "Qwen/Qwen3-8B", "Qwen/Qwen3-4B", "Qwen/Qwen3-32B",
+        "Qwen/Qwen2.5-7B-Instruct-1M", "meta-llama/Llama-3.2-3B-Instruct"]
     seen = set()
     for m in cand:
         if not m or m in seen:
@@ -548,7 +550,7 @@ async def api_ai_probe():
             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]})
+            out.append({"model": m, "ok": False, "sec": round(_t.time() - t0, 1), "err": (type(e).__name__ + ": " + str(e))[-300:]})
     return JSONResponse({"results": out})