Complete local hybrid search and improve agent reliability

2026-10-02 23:38:02 +03:00
parent 3563a104a3
commit 140f6eb287
62 changed files with 7546 additions and 314 deletions
@@ -0,0 +1,46 @@
+{
+  "created_at_utc": "2026-10-02T20:30:24.674307+00:00",
+  "base_url": "http://127.0.0.1:8002",
+  "requested_model": "qwen2.5:1.5b-instruct-q4_K_M",
+  "dataset": "evals\\baseline_arabic_programming.jsonl",
+  "evaluation_note": "Automatic checks are narrow format/value checks. Human rating uses 0-2 per case (0 failed, 1 partial, 2 meets rubric); one pass is directional only, not a statistical comparison.",
+  "warmup": null,
+  "latency_summary_seconds": {
+    "mean": 21.85,
+    "median": 21.85,
+    "min": 21.85,
+    "max": 21.85
+  },
+  "results": [
+    {
+      "id": "arithmetic-tool",
+      "category": "tool-use",
+      "rubric": [
+        "استُدعيت أداة الحاسبة عند توفرها",
+        "الناتج 3973 دون ادعاء غير صحيح"
+      ],
+      "status": "ok",
+      "elapsed_seconds": 21.85,
+      "answer": "الناتج هو 3973.",
+      "tool": "calculator",
+      "checks": [
+        {
+          "check": {
+            "type": "contains",
+            "value": "3973"
+          },
+          "passed": true
+        },
+        {
+          "check": {
+            "type": "tool_is",
+            "value": "calculator"
+          },
+          "passed": true
+        }
+      ],
+      "human_rating": null,
+      "human_notes": ""
+    }
+  ]
+}