Training checkpoint at step 4000
Browse files- trainer_state.json +366 -6
trainer_state.json
CHANGED
|
@@ -1,10 +1,10 @@
|
|
| 1 |
{
|
| 2 |
-
"best_global_step":
|
| 3 |
-
"best_metric": 2.
|
| 4 |
-
"best_model_checkpoint": "../runs/karpathy/fineweb-edu-100b-shuffle/meta-llama/Llama-3.2-1B/linear_adamw_wd1e-03_7x1024_mem32_bs64_hf_armt_dmem64/run_30/checkpoint-
|
| 5 |
-
"epoch": 0.
|
| 6 |
"eval_steps": 100,
|
| 7 |
-
"global_step":
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
|
@@ -1088,6 +1088,366 @@
|
|
| 1088 |
"eval_samples_per_second": 3.517,
|
| 1089 |
"eval_steps_per_second": 1.773,
|
| 1090 |
"step": 3000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1091 |
}
|
| 1092 |
],
|
| 1093 |
"logging_steps": 25,
|
|
@@ -1107,7 +1467,7 @@
|
|
| 1107 |
"attributes": {}
|
| 1108 |
}
|
| 1109 |
},
|
| 1110 |
-
"total_flos":
|
| 1111 |
"train_batch_size": 1,
|
| 1112 |
"trial_name": null,
|
| 1113 |
"trial_params": null
|
|
|
|
| 1 |
{
|
| 2 |
+
"best_global_step": 4000,
|
| 3 |
+
"best_metric": 2.430954933166504,
|
| 4 |
+
"best_model_checkpoint": "../runs/karpathy/fineweb-edu-100b-shuffle/meta-llama/Llama-3.2-1B/linear_adamw_wd1e-03_7x1024_mem32_bs64_hf_armt_dmem64/run_30/checkpoint-4000",
|
| 5 |
+
"epoch": 0.08,
|
| 6 |
"eval_steps": 100,
|
| 7 |
+
"global_step": 4000,
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
|
|
|
| 1088 |
"eval_samples_per_second": 3.517,
|
| 1089 |
"eval_steps_per_second": 1.773,
|
| 1090 |
"step": 3000
|
| 1091 |
+
},
|
| 1092 |
+
{
|
| 1093 |
+
"epoch": 0.0605,
|
| 1094 |
+
"grad_norm": 0.025507916252978883,
|
| 1095 |
+
"learning_rate": 6.048e-06,
|
| 1096 |
+
"loss": 2.4192,
|
| 1097 |
+
"step": 3025
|
| 1098 |
+
},
|
| 1099 |
+
{
|
| 1100 |
+
"epoch": 0.061,
|
| 1101 |
+
"grad_norm": 0.02126046028834697,
|
| 1102 |
+
"learning_rate": 6.098000000000001e-06,
|
| 1103 |
+
"loss": 2.4233,
|
| 1104 |
+
"step": 3050
|
| 1105 |
+
},
|
| 1106 |
+
{
|
| 1107 |
+
"epoch": 0.0615,
|
| 1108 |
+
"grad_norm": 0.026235681014214807,
|
| 1109 |
+
"learning_rate": 6.148e-06,
|
| 1110 |
+
"loss": 2.4215,
|
| 1111 |
+
"step": 3075
|
| 1112 |
+
},
|
| 1113 |
+
{
|
| 1114 |
+
"epoch": 0.062,
|
| 1115 |
+
"grad_norm": 0.026243691288249413,
|
| 1116 |
+
"learning_rate": 6.198000000000001e-06,
|
| 1117 |
+
"loss": 2.4134,
|
| 1118 |
+
"step": 3100
|
| 1119 |
+
},
|
| 1120 |
+
{
|
| 1121 |
+
"epoch": 0.062,
|
| 1122 |
+
"eval_loss": 2.435988664627075,
|
| 1123 |
+
"eval_runtime": 33.0276,
|
| 1124 |
+
"eval_samples_per_second": 3.542,
|
| 1125 |
+
"eval_steps_per_second": 1.786,
|
| 1126 |
+
"step": 3100
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 0.0625,
|
| 1130 |
+
"grad_norm": 0.02496599291141367,
|
| 1131 |
+
"learning_rate": 6.248000000000001e-06,
|
| 1132 |
+
"loss": 2.4241,
|
| 1133 |
+
"step": 3125
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 0.063,
|
| 1137 |
+
"grad_norm": 0.0236951365360608,
|
| 1138 |
+
"learning_rate": 6.2980000000000005e-06,
|
| 1139 |
+
"loss": 2.4252,
|
| 1140 |
+
"step": 3150
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 0.0635,
|
| 1144 |
+
"grad_norm": 0.022752035914773892,
|
| 1145 |
+
"learning_rate": 6.348000000000001e-06,
|
| 1146 |
+
"loss": 2.4244,
|
| 1147 |
+
"step": 3175
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 0.064,
|
| 1151 |
+
"grad_norm": 0.021656953860252137,
|
| 1152 |
+
"learning_rate": 6.398000000000001e-06,
|
| 1153 |
+
"loss": 2.4227,
|
| 1154 |
+
"step": 3200
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 0.064,
|
| 1158 |
+
"eval_loss": 2.43520450592041,
|
| 1159 |
+
"eval_runtime": 33.136,
|
| 1160 |
+
"eval_samples_per_second": 3.531,
|
| 1161 |
+
"eval_steps_per_second": 1.781,
|
| 1162 |
+
"step": 3200
|
| 1163 |
+
},
|
| 1164 |
+
{
|
| 1165 |
+
"epoch": 0.0645,
|
| 1166 |
+
"grad_norm": 0.021188520683488872,
|
| 1167 |
+
"learning_rate": 6.448000000000001e-06,
|
| 1168 |
+
"loss": 2.4248,
|
| 1169 |
+
"step": 3225
|
| 1170 |
+
},
|
| 1171 |
+
{
|
| 1172 |
+
"epoch": 0.065,
|
| 1173 |
+
"grad_norm": 0.02274972468402099,
|
| 1174 |
+
"learning_rate": 6.498000000000001e-06,
|
| 1175 |
+
"loss": 2.4215,
|
| 1176 |
+
"step": 3250
|
| 1177 |
+
},
|
| 1178 |
+
{
|
| 1179 |
+
"epoch": 0.0655,
|
| 1180 |
+
"grad_norm": 0.024046700552500286,
|
| 1181 |
+
"learning_rate": 6.548000000000001e-06,
|
| 1182 |
+
"loss": 2.4169,
|
| 1183 |
+
"step": 3275
|
| 1184 |
+
},
|
| 1185 |
+
{
|
| 1186 |
+
"epoch": 0.066,
|
| 1187 |
+
"grad_norm": 0.022071385618052216,
|
| 1188 |
+
"learning_rate": 6.598000000000001e-06,
|
| 1189 |
+
"loss": 2.4199,
|
| 1190 |
+
"step": 3300
|
| 1191 |
+
},
|
| 1192 |
+
{
|
| 1193 |
+
"epoch": 0.066,
|
| 1194 |
+
"eval_loss": 2.4344840049743652,
|
| 1195 |
+
"eval_runtime": 33.1729,
|
| 1196 |
+
"eval_samples_per_second": 3.527,
|
| 1197 |
+
"eval_steps_per_second": 1.779,
|
| 1198 |
+
"step": 3300
|
| 1199 |
+
},
|
| 1200 |
+
{
|
| 1201 |
+
"epoch": 0.0665,
|
| 1202 |
+
"grad_norm": 0.02931021842271797,
|
| 1203 |
+
"learning_rate": 6.648e-06,
|
| 1204 |
+
"loss": 2.4253,
|
| 1205 |
+
"step": 3325
|
| 1206 |
+
},
|
| 1207 |
+
{
|
| 1208 |
+
"epoch": 0.067,
|
| 1209 |
+
"grad_norm": 0.021754527434557868,
|
| 1210 |
+
"learning_rate": 6.698e-06,
|
| 1211 |
+
"loss": 2.4281,
|
| 1212 |
+
"step": 3350
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 0.0675,
|
| 1216 |
+
"grad_norm": 0.022651522972508432,
|
| 1217 |
+
"learning_rate": 6.7480000000000004e-06,
|
| 1218 |
+
"loss": 2.4208,
|
| 1219 |
+
"step": 3375
|
| 1220 |
+
},
|
| 1221 |
+
{
|
| 1222 |
+
"epoch": 0.068,
|
| 1223 |
+
"grad_norm": 0.022676405563792287,
|
| 1224 |
+
"learning_rate": 6.798e-06,
|
| 1225 |
+
"loss": 2.4222,
|
| 1226 |
+
"step": 3400
|
| 1227 |
+
},
|
| 1228 |
+
{
|
| 1229 |
+
"epoch": 0.068,
|
| 1230 |
+
"eval_loss": 2.43371844291687,
|
| 1231 |
+
"eval_runtime": 33.1293,
|
| 1232 |
+
"eval_samples_per_second": 3.532,
|
| 1233 |
+
"eval_steps_per_second": 1.781,
|
| 1234 |
+
"step": 3400
|
| 1235 |
+
},
|
| 1236 |
+
{
|
| 1237 |
+
"epoch": 0.0685,
|
| 1238 |
+
"grad_norm": 0.021100680573628707,
|
| 1239 |
+
"learning_rate": 6.848e-06,
|
| 1240 |
+
"loss": 2.4243,
|
| 1241 |
+
"step": 3425
|
| 1242 |
+
},
|
| 1243 |
+
{
|
| 1244 |
+
"epoch": 0.069,
|
| 1245 |
+
"grad_norm": 0.02101417038408147,
|
| 1246 |
+
"learning_rate": 6.898e-06,
|
| 1247 |
+
"loss": 2.4242,
|
| 1248 |
+
"step": 3450
|
| 1249 |
+
},
|
| 1250 |
+
{
|
| 1251 |
+
"epoch": 0.0695,
|
| 1252 |
+
"grad_norm": 0.022432735157488455,
|
| 1253 |
+
"learning_rate": 6.948e-06,
|
| 1254 |
+
"loss": 2.4224,
|
| 1255 |
+
"step": 3475
|
| 1256 |
+
},
|
| 1257 |
+
{
|
| 1258 |
+
"epoch": 0.07,
|
| 1259 |
+
"grad_norm": 0.02164716008760555,
|
| 1260 |
+
"learning_rate": 6.998000000000001e-06,
|
| 1261 |
+
"loss": 2.4202,
|
| 1262 |
+
"step": 3500
|
| 1263 |
+
},
|
| 1264 |
+
{
|
| 1265 |
+
"epoch": 0.07,
|
| 1266 |
+
"eval_loss": 2.433281898498535,
|
| 1267 |
+
"eval_runtime": 33.0783,
|
| 1268 |
+
"eval_samples_per_second": 3.537,
|
| 1269 |
+
"eval_steps_per_second": 1.784,
|
| 1270 |
+
"step": 3500
|
| 1271 |
+
},
|
| 1272 |
+
{
|
| 1273 |
+
"epoch": 0.0705,
|
| 1274 |
+
"grad_norm": 0.022412840176404082,
|
| 1275 |
+
"learning_rate": 7.048e-06,
|
| 1276 |
+
"loss": 2.4184,
|
| 1277 |
+
"step": 3525
|
| 1278 |
+
},
|
| 1279 |
+
{
|
| 1280 |
+
"epoch": 0.071,
|
| 1281 |
+
"grad_norm": 0.025300113537910857,
|
| 1282 |
+
"learning_rate": 7.0980000000000005e-06,
|
| 1283 |
+
"loss": 2.421,
|
| 1284 |
+
"step": 3550
|
| 1285 |
+
},
|
| 1286 |
+
{
|
| 1287 |
+
"epoch": 0.0715,
|
| 1288 |
+
"grad_norm": 0.022085711512698558,
|
| 1289 |
+
"learning_rate": 7.148000000000001e-06,
|
| 1290 |
+
"loss": 2.415,
|
| 1291 |
+
"step": 3575
|
| 1292 |
+
},
|
| 1293 |
+
{
|
| 1294 |
+
"epoch": 0.072,
|
| 1295 |
+
"grad_norm": 0.021041258769866313,
|
| 1296 |
+
"learning_rate": 7.198e-06,
|
| 1297 |
+
"loss": 2.4157,
|
| 1298 |
+
"step": 3600
|
| 1299 |
+
},
|
| 1300 |
+
{
|
| 1301 |
+
"epoch": 0.072,
|
| 1302 |
+
"eval_loss": 2.4324123859405518,
|
| 1303 |
+
"eval_runtime": 34.1633,
|
| 1304 |
+
"eval_samples_per_second": 3.425,
|
| 1305 |
+
"eval_steps_per_second": 1.727,
|
| 1306 |
+
"step": 3600
|
| 1307 |
+
},
|
| 1308 |
+
{
|
| 1309 |
+
"epoch": 0.0725,
|
| 1310 |
+
"grad_norm": 0.021694681795354324,
|
| 1311 |
+
"learning_rate": 7.248000000000001e-06,
|
| 1312 |
+
"loss": 2.4152,
|
| 1313 |
+
"step": 3625
|
| 1314 |
+
},
|
| 1315 |
+
{
|
| 1316 |
+
"epoch": 0.073,
|
| 1317 |
+
"grad_norm": 0.03056130171104773,
|
| 1318 |
+
"learning_rate": 7.298e-06,
|
| 1319 |
+
"loss": 2.4151,
|
| 1320 |
+
"step": 3650
|
| 1321 |
+
},
|
| 1322 |
+
{
|
| 1323 |
+
"epoch": 0.0735,
|
| 1324 |
+
"grad_norm": 0.02112814663770162,
|
| 1325 |
+
"learning_rate": 7.348000000000001e-06,
|
| 1326 |
+
"loss": 2.4163,
|
| 1327 |
+
"step": 3675
|
| 1328 |
+
},
|
| 1329 |
+
{
|
| 1330 |
+
"epoch": 0.074,
|
| 1331 |
+
"grad_norm": 0.024883267721069864,
|
| 1332 |
+
"learning_rate": 7.398000000000001e-06,
|
| 1333 |
+
"loss": 2.4258,
|
| 1334 |
+
"step": 3700
|
| 1335 |
+
},
|
| 1336 |
+
{
|
| 1337 |
+
"epoch": 0.074,
|
| 1338 |
+
"eval_loss": 2.4319984912872314,
|
| 1339 |
+
"eval_runtime": 33.2699,
|
| 1340 |
+
"eval_samples_per_second": 3.517,
|
| 1341 |
+
"eval_steps_per_second": 1.773,
|
| 1342 |
+
"step": 3700
|
| 1343 |
+
},
|
| 1344 |
+
{
|
| 1345 |
+
"epoch": 0.0745,
|
| 1346 |
+
"grad_norm": 0.02062910451612879,
|
| 1347 |
+
"learning_rate": 7.4480000000000005e-06,
|
| 1348 |
+
"loss": 2.4263,
|
| 1349 |
+
"step": 3725
|
| 1350 |
+
},
|
| 1351 |
+
{
|
| 1352 |
+
"epoch": 0.075,
|
| 1353 |
+
"grad_norm": 0.021068085012478772,
|
| 1354 |
+
"learning_rate": 7.498000000000001e-06,
|
| 1355 |
+
"loss": 2.4216,
|
| 1356 |
+
"step": 3750
|
| 1357 |
+
},
|
| 1358 |
+
{
|
| 1359 |
+
"epoch": 0.0755,
|
| 1360 |
+
"grad_norm": 0.020665118516629687,
|
| 1361 |
+
"learning_rate": 7.548000000000001e-06,
|
| 1362 |
+
"loss": 2.4285,
|
| 1363 |
+
"step": 3775
|
| 1364 |
+
},
|
| 1365 |
+
{
|
| 1366 |
+
"epoch": 0.076,
|
| 1367 |
+
"grad_norm": 0.02425992757924128,
|
| 1368 |
+
"learning_rate": 7.598000000000001e-06,
|
| 1369 |
+
"loss": 2.4174,
|
| 1370 |
+
"step": 3800
|
| 1371 |
+
},
|
| 1372 |
+
{
|
| 1373 |
+
"epoch": 0.076,
|
| 1374 |
+
"eval_loss": 2.4310436248779297,
|
| 1375 |
+
"eval_runtime": 35.0728,
|
| 1376 |
+
"eval_samples_per_second": 3.336,
|
| 1377 |
+
"eval_steps_per_second": 1.682,
|
| 1378 |
+
"step": 3800
|
| 1379 |
+
},
|
| 1380 |
+
{
|
| 1381 |
+
"epoch": 0.0765,
|
| 1382 |
+
"grad_norm": 0.021337004595007786,
|
| 1383 |
+
"learning_rate": 7.648e-06,
|
| 1384 |
+
"loss": 2.4303,
|
| 1385 |
+
"step": 3825
|
| 1386 |
+
},
|
| 1387 |
+
{
|
| 1388 |
+
"epoch": 0.077,
|
| 1389 |
+
"grad_norm": 0.020168500131750186,
|
| 1390 |
+
"learning_rate": 7.698000000000002e-06,
|
| 1391 |
+
"loss": 2.4298,
|
| 1392 |
+
"step": 3850
|
| 1393 |
+
},
|
| 1394 |
+
{
|
| 1395 |
+
"epoch": 0.0775,
|
| 1396 |
+
"grad_norm": 0.020089032493824672,
|
| 1397 |
+
"learning_rate": 7.748000000000001e-06,
|
| 1398 |
+
"loss": 2.4151,
|
| 1399 |
+
"step": 3875
|
| 1400 |
+
},
|
| 1401 |
+
{
|
| 1402 |
+
"epoch": 0.078,
|
| 1403 |
+
"grad_norm": 0.02462630071931115,
|
| 1404 |
+
"learning_rate": 7.798e-06,
|
| 1405 |
+
"loss": 2.4235,
|
| 1406 |
+
"step": 3900
|
| 1407 |
+
},
|
| 1408 |
+
{
|
| 1409 |
+
"epoch": 0.078,
|
| 1410 |
+
"eval_loss": 2.431330442428589,
|
| 1411 |
+
"eval_runtime": 33.093,
|
| 1412 |
+
"eval_samples_per_second": 3.535,
|
| 1413 |
+
"eval_steps_per_second": 1.783,
|
| 1414 |
+
"step": 3900
|
| 1415 |
+
},
|
| 1416 |
+
{
|
| 1417 |
+
"epoch": 0.0785,
|
| 1418 |
+
"grad_norm": 0.0226705620922379,
|
| 1419 |
+
"learning_rate": 7.848000000000002e-06,
|
| 1420 |
+
"loss": 2.4185,
|
| 1421 |
+
"step": 3925
|
| 1422 |
+
},
|
| 1423 |
+
{
|
| 1424 |
+
"epoch": 0.079,
|
| 1425 |
+
"grad_norm": 0.022075041269811142,
|
| 1426 |
+
"learning_rate": 7.898e-06,
|
| 1427 |
+
"loss": 2.4344,
|
| 1428 |
+
"step": 3950
|
| 1429 |
+
},
|
| 1430 |
+
{
|
| 1431 |
+
"epoch": 0.0795,
|
| 1432 |
+
"grad_norm": 0.03932607113814955,
|
| 1433 |
+
"learning_rate": 7.948e-06,
|
| 1434 |
+
"loss": 2.4228,
|
| 1435 |
+
"step": 3975
|
| 1436 |
+
},
|
| 1437 |
+
{
|
| 1438 |
+
"epoch": 0.08,
|
| 1439 |
+
"grad_norm": 0.020604342831921824,
|
| 1440 |
+
"learning_rate": 7.998e-06,
|
| 1441 |
+
"loss": 2.4289,
|
| 1442 |
+
"step": 4000
|
| 1443 |
+
},
|
| 1444 |
+
{
|
| 1445 |
+
"epoch": 0.08,
|
| 1446 |
+
"eval_loss": 2.430954933166504,
|
| 1447 |
+
"eval_runtime": 33.1216,
|
| 1448 |
+
"eval_samples_per_second": 3.532,
|
| 1449 |
+
"eval_steps_per_second": 1.781,
|
| 1450 |
+
"step": 4000
|
| 1451 |
}
|
| 1452 |
],
|
| 1453 |
"logging_steps": 25,
|
|
|
|
| 1467 |
"attributes": {}
|
| 1468 |
}
|
| 1469 |
},
|
| 1470 |
+
"total_flos": 1.114120770054783e+19,
|
| 1471 |
"train_batch_size": 1,
|
| 1472 |
"trial_name": null,
|
| 1473 |
"trial_params": null
|