zkas-1B: architecture-controlled twin — results only, no weights
Browse files- results.json +164 -0
results.json
ADDED
|
@@ -0,0 +1,164 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "tetracta_zkas_1B",
|
| 3 |
+
"params": 953466904,
|
| 4 |
+
"pretraining": {
|
| 5 |
+
"steps": 315000,
|
| 6 |
+
"tokens": 10321920000,
|
| 7 |
+
"epochs": 1,
|
| 8 |
+
"skipped_steps": 0,
|
| 9 |
+
"val_bpc_raw": 4.8571,
|
| 10 |
+
"val_bpc_clean_7of8": 4.3929,
|
| 11 |
+
"val_ppl_char_clean": 21.01,
|
| 12 |
+
"note": "one eval sequence of 8 diverged pathologically; both numbers reported"
|
| 13 |
+
},
|
| 14 |
+
"sft": {
|
| 15 |
+
"examples": 11419,
|
| 16 |
+
"epochs": 3,
|
| 17 |
+
"lr": 1e-05,
|
| 18 |
+
"steps": 2142,
|
| 19 |
+
"seconds": 1247,
|
| 20 |
+
"skipped_steps": 0,
|
| 21 |
+
"identical_to": "tetracta_van_1b"
|
| 22 |
+
},
|
| 23 |
+
"xray": {
|
| 24 |
+
"zkas_taban": {
|
| 25 |
+
"agirlik_kaymasi": null,
|
| 26 |
+
"duz_metin_kayip": 3.4415,
|
| 27 |
+
"duz_metin_bpc_yaklasik": 4.9651,
|
| 28 |
+
"logit_entropi": 7.8817,
|
| 29 |
+
"tepe_olasilik": 0.0961,
|
| 30 |
+
"ozel_token_olasiligi": 1.6e-05
|
| 31 |
+
},
|
| 32 |
+
"zsft": {
|
| 33 |
+
"agirlik_kaymasi": {
|
| 34 |
+
"Win": 0.01119,
|
| 35 |
+
"Wo": 0.01299,
|
| 36 |
+
"alfa": 0.0256,
|
| 37 |
+
"c0": 0.0021,
|
| 38 |
+
"c_katsayi": 0.0007,
|
| 39 |
+
"emb": 0.05327,
|
| 40 |
+
"sayac_Wf": 0.18723,
|
| 41 |
+
"sayac_Wg": 0.00551,
|
| 42 |
+
"theta_gama": 9e-05
|
| 43 |
+
},
|
| 44 |
+
"kayma_katman": [
|
| 45 |
+
0.00357,
|
| 46 |
+
0.00363,
|
| 47 |
+
0.00405,
|
| 48 |
+
0.00429,
|
| 49 |
+
0.00425,
|
| 50 |
+
0.00424,
|
| 51 |
+
0.00447,
|
| 52 |
+
0.00462,
|
| 53 |
+
0.0047,
|
| 54 |
+
0.00464,
|
| 55 |
+
0.00457,
|
| 56 |
+
0.00462,
|
| 57 |
+
0.00473,
|
| 58 |
+
0.00484,
|
| 59 |
+
0.00504,
|
| 60 |
+
0.00506,
|
| 61 |
+
0.00516,
|
| 62 |
+
0.00524,
|
| 63 |
+
0.00543,
|
| 64 |
+
0.00606,
|
| 65 |
+
0.00648,
|
| 66 |
+
0.01235,
|
| 67 |
+
0.0325,
|
| 68 |
+
0.06618
|
| 69 |
+
],
|
| 70 |
+
"kayma_toplam": 0.01056,
|
| 71 |
+
"duz_metin_kayip": 4.6756,
|
| 72 |
+
"duz_metin_bpc_yaklasik": 6.7455,
|
| 73 |
+
"logit_entropi": 1.6244,
|
| 74 |
+
"tepe_olasilik": 0.7221,
|
| 75 |
+
"ozel_token_olasiligi": 1.6e-05
|
| 76 |
+
}
|
| 77 |
+
},
|
| 78 |
+
"judge_panel": {
|
| 79 |
+
"qwen15base": {
|
| 80 |
+
"talimat": {
|
| 81 |
+
"zsft": 1,
|
| 82 |
+
"rakip": 7,
|
| 83 |
+
"berabere": 0
|
| 84 |
+
},
|
| 85 |
+
"yazma": {
|
| 86 |
+
"zsft": 0,
|
| 87 |
+
"rakip": 8,
|
| 88 |
+
"berabere": 0
|
| 89 |
+
},
|
| 90 |
+
"bilgi": {
|
| 91 |
+
"zsft": 1,
|
| 92 |
+
"rakip": 7,
|
| 93 |
+
"berabere": 0
|
| 94 |
+
},
|
| 95 |
+
"sohbet": {
|
| 96 |
+
"zsft": 6,
|
| 97 |
+
"rakip": 2,
|
| 98 |
+
"berabere": 0
|
| 99 |
+
},
|
| 100 |
+
"muhakeme": {
|
| 101 |
+
"zsft": 0,
|
| 102 |
+
"rakip": 8,
|
| 103 |
+
"berabere": 0
|
| 104 |
+
},
|
| 105 |
+
"guvenlik": {
|
| 106 |
+
"zsft": 4,
|
| 107 |
+
"rakip": 4,
|
| 108 |
+
"berabere": 0
|
| 109 |
+
},
|
| 110 |
+
"durustluk": {
|
| 111 |
+
"zsft": 5,
|
| 112 |
+
"rakip": 3,
|
| 113 |
+
"berabere": 0
|
| 114 |
+
},
|
| 115 |
+
"kimlik": {
|
| 116 |
+
"zsft": 7,
|
| 117 |
+
"rakip": 0,
|
| 118 |
+
"berabere": 1
|
| 119 |
+
}
|
| 120 |
+
},
|
| 121 |
+
"van_sft2": {
|
| 122 |
+
"bilgi": {
|
| 123 |
+
"zsft": 4,
|
| 124 |
+
"rakip": 3,
|
| 125 |
+
"berabere": 1
|
| 126 |
+
},
|
| 127 |
+
"durustluk": {
|
| 128 |
+
"zsft": 1,
|
| 129 |
+
"rakip": 7,
|
| 130 |
+
"berabere": 0
|
| 131 |
+
},
|
| 132 |
+
"talimat": {
|
| 133 |
+
"zsft": 0,
|
| 134 |
+
"rakip": 7,
|
| 135 |
+
"berabere": 1
|
| 136 |
+
},
|
| 137 |
+
"guvenlik": {
|
| 138 |
+
"zsft": 1,
|
| 139 |
+
"rakip": 7,
|
| 140 |
+
"berabere": 0
|
| 141 |
+
},
|
| 142 |
+
"kimlik": {
|
| 143 |
+
"zsft": 1,
|
| 144 |
+
"rakip": 6,
|
| 145 |
+
"berabere": 1
|
| 146 |
+
},
|
| 147 |
+
"muhakeme": {
|
| 148 |
+
"zsft": 2,
|
| 149 |
+
"rakip": 5,
|
| 150 |
+
"berabere": 1
|
| 151 |
+
},
|
| 152 |
+
"sohbet": {
|
| 153 |
+
"zsft": 3,
|
| 154 |
+
"rakip": 5,
|
| 155 |
+
"berabere": 0
|
| 156 |
+
},
|
| 157 |
+
"yazma": {
|
| 158 |
+
"zsft": 3,
|
| 159 |
+
"rakip": 5,
|
| 160 |
+
"berabere": 0
|
| 161 |
+
}
|
| 162 |
+
}
|
| 163 |
+
}
|
| 164 |
+
}
|