Update benchmarks/mistral-small4-119b-sft-sft-dpo (20261007-082142-332282) (part 2)

benchmarks/mistral-small4-119b-sft-sft-dpo/raw/toxicity.json

@@ -0,0 +1,16 @@
+{
+ "name": "toxicity",
+ "kind": "toxicity",
+ "group": "toxicity",
+ "params": {
+  "num_samples": 400
+ },
+ "cost": 3200,
+ "metrics": {
+  "mean_toxicity": 0.10175457717024983,
+  "std_toxicity": 0.2533916548542065,
+  "n": 400
+ },
+ "seconds": 11.5,
+ "model_load_seconds": 137.8
+}

benchmarks/mistral-small4-119b-sft-sft-dpo/summary.json

@@ -1,51 +1,15 @@
 {
  "model_id": "models/mistral_small4-sft-sft-dpo-native",
  "summary": {
-  "kmmlu_direct": {
-   "exact_match,none": 0.6302026834142164,
-   "n": 35030,
-   "complete": true,
-   "by_category": {
-    "applied_science": {
-     "exact_match,none": 0.6217241379310345,
-     "n": 11600
-    },
-    "humss": {
-     "exact_match,none": 0.6479532163742691,
-     "n": 5130
-    },
-    "other": {
-     "exact_match,none": 0.5991666666666666,
-     "n": 8400
-    },
-    "stem": {
-     "exact_match,none": 0.6572727272727272,
-     "n": 9900
-    }
-   }
-  },
-  "ko_hellaswag": {
-   "acc,none": 0.4671380203146783,
-   "acc_norm,none": 0.6175064728141805,
-   "n": 10042,
-   "complete": true
-  },
-  "ko_truthfulqa_mc2": {
-   "acc,none": 0.5148032721223216,
-   "n": 817,
-   "complete": true
-  },
-  "emotion": {
-   "accuracy": 0.6574712643678161,
-   "accuracy_normalized": 0.6574712643678161,
-   "n": 2610,
+  "toxicity": {
+   "mean_toxicity": 0.10175457717024983,
+   "std_toxicity": 0.2533916548542065,
+   "n": 400,
    "complete": true
   }
  },
- "missing": [
-  "toxicity"
- ],
- "wall_seconds": 889.3,
+ "missing": [],
+ "wall_seconds": 163.7,
  "plan": {
   "tp": 4,
   "gpus": "0,1,2,3,4,5,6,7"

benchmarks/mistral-small4-119b-sft-sft-dpo/summary.md

@@ -1,21 +1,13 @@
 # Benchmark: models/mistral_small4-sft-sft-dpo-native
 
-- 실행 시각: 2026-10-07 08:14:45
-- GPU 슬롯: 2개 x tp=4 (GPU 0,1,2,3,4,5,6,7)
-- 전체 소요(wall-clock): 14.8분
+- 실행 시각: 2026-10-07 08:24:29
+- GPU 슬롯: 1개 x tp=4 (GPU 0,1,2,3,4,5,6,7)
+- 전체 소요(wall-clock): 2.7분
 
 | 벤치마크 | 점수 | 문항 수 | 완료 |
 |---|---|---:|---|
-| ko_truthfulqa_mc2 | acc 0.5148 | 817 | O |
-| ko_hellaswag | acc 0.4671, acc_norm 0.6175 | 10,042 | O |
-| kmmlu_direct | exact_match 0.6302 | 35,030 | O |
-| emotion | accuracy 0.6575, accuracy_normalized 0.6575 | 2,610 | O |
-
-KMMLU-direct 카테고리별 exact_match: applied_science 0.6217, humss 0.6480, other 0.5992, stem 0.6573
+| toxicity | mean_toxicity 0.1018, std_toxicity 0.2534 | 400 | O |
 
 | 슬롯 | GPU | 작업 수 | 모델 로드(s) | 작업 합계(s) |
 |---|---|---:|---:|---:|
-| 0 | 0,1,2,3 | 28 | 152 | 720 |
-| 1 | 4,5,6,7 | 28 | 153 | 692 |
-
-**미완료 작업 1개**: toxicity (같은 명령을 다시 실행하면 이어서 진행)
+| 0 | 0,1,2,3 | 1 | 138 | 12 |