Add full V5.9.2 evaluation artifacts
Browse files
eval/significance_tests_v592.json
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"27B_AdamW_step400_vs_9B_AdamW_step200": {
|
| 3 |
+
"mean_diff": 4.896,
|
| 4 |
+
"bootstrap_ci95": [
|
| 5 |
+
2.867,
|
| 6 |
+
6.789
|
| 7 |
+
],
|
| 8 |
+
"wilcoxon_W": 5719.5,
|
| 9 |
+
"wilcoxon_p": 4.847597288453082e-07,
|
| 10 |
+
"significant_at_0.05": true
|
| 11 |
+
},
|
| 12 |
+
"27B_AdamW_step400_vs_4B_AdamW_step200": {
|
| 13 |
+
"mean_diff": 6.978,
|
| 14 |
+
"bootstrap_ci95": [
|
| 15 |
+
4.814,
|
| 16 |
+
9.154
|
| 17 |
+
],
|
| 18 |
+
"wilcoxon_W": 4877.0,
|
| 19 |
+
"wilcoxon_p": 4.47399700563063e-10,
|
| 20 |
+
"significant_at_0.05": true
|
| 21 |
+
},
|
| 22 |
+
"27B_AdamW_step400_vs_4B_Muon_step200": {
|
| 23 |
+
"mean_diff": 15.704,
|
| 24 |
+
"bootstrap_ci95": [
|
| 25 |
+
13.551,
|
| 26 |
+
17.92
|
| 27 |
+
],
|
| 28 |
+
"wilcoxon_W": 1574.5,
|
| 29 |
+
"wilcoxon_p": 4.570449701547677e-25,
|
| 30 |
+
"significant_at_0.05": true
|
| 31 |
+
},
|
| 32 |
+
"9B_AdamW_step200_vs_4B_AdamW_step200": {
|
| 33 |
+
"mean_diff": 2.082,
|
| 34 |
+
"bootstrap_ci95": [
|
| 35 |
+
-0.034,
|
| 36 |
+
4.176
|
| 37 |
+
],
|
| 38 |
+
"wilcoxon_W": 8359.5,
|
| 39 |
+
"wilcoxon_p": 0.0647704989672306,
|
| 40 |
+
"significant_at_0.05": false
|
| 41 |
+
},
|
| 42 |
+
"9B_AdamW_step200_vs_4B_Muon_step200": {
|
| 43 |
+
"mean_diff": 10.808,
|
| 44 |
+
"bootstrap_ci95": [
|
| 45 |
+
8.818,
|
| 46 |
+
12.804
|
| 47 |
+
],
|
| 48 |
+
"wilcoxon_W": 2929.0,
|
| 49 |
+
"wilcoxon_p": 3.661798966685315e-18,
|
| 50 |
+
"significant_at_0.05": true
|
| 51 |
+
},
|
| 52 |
+
"4B_AdamW_step200_vs_4B_Muon_step200": {
|
| 53 |
+
"mean_diff": 8.726,
|
| 54 |
+
"bootstrap_ci95": [
|
| 55 |
+
6.512,
|
| 56 |
+
10.939
|
| 57 |
+
],
|
| 58 |
+
"wilcoxon_W": 4134.5,
|
| 59 |
+
"wilcoxon_p": 5.27884641909125e-13,
|
| 60 |
+
"significant_at_0.05": true
|
| 61 |
+
}
|
| 62 |
+
}
|