sharukat commited on
Commit
9e4720b
·
verified ·
1 Parent(s): a9bd34d

Training in progress, epoch 5

Browse files
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:68f0e33a448107c1f1b74af262a7d7f972740eadf4fb074a5cf5f6d632989485
3
  size 502675828
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b4fc6a310fdfbb69d44f9cb70c3ea69194d45369bbd141dca0a43b4272a0ba46
3
  size 502675828
runs/Mar07_03-35-46_8bb175de6f67/events.out.tfevents.1709782548.8bb175de6f67.34.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:500b87b7d3ea29737aeb31d97332e2ca0f9fc47513f133d1333f197bf069387e
3
- size 7873
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ba6fdc7e588f9e0cb99c074c7d159b3489543982bdb8b41d625f2bc75febf3e4
3
+ size 8556
wandb/debug-internal.log CHANGED
@@ -181,3 +181,39 @@
181
  2024-03-07 03:39:03,990 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
182
  2024-03-07 03:39:05,518 DEBUG SenderThread:147 [sender.py:send():382] send: stats
183
  2024-03-07 03:39:06,291 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
181
  2024-03-07 03:39:03,990 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
182
  2024-03-07 03:39:05,518 DEBUG SenderThread:147 [sender.py:send():382] send: stats
183
  2024-03-07 03:39:06,291 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
184
+ 2024-03-07 03:39:09,519 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
185
+ 2024-03-07 03:39:11,292 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
186
+ 2024-03-07 03:39:14,520 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
187
+ 2024-03-07 03:39:16,294 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
188
+ 2024-03-07 03:39:19,521 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
189
+ 2024-03-07 03:39:21,295 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
190
+ 2024-03-07 03:39:22,389 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: partial_history
191
+ 2024-03-07 03:39:22,390 DEBUG SenderThread:147 [sender.py:send():382] send: history
192
+ 2024-03-07 03:39:22,390 DEBUG SenderThread:147 [sender.py:send_request():409] send_request: summary_record
193
+ 2024-03-07 03:39:22,391 INFO SenderThread:147 [sender.py:_save_file():1403] saving file wandb-summary.json with policy end
194
+ 2024-03-07 03:39:22,513 INFO Thread-12 :147 [dir_watcher.py:_on_file_modified():288] file/dir modified: /kaggle/working/wandb/run-20240307_033604-u9myy8y3/files/wandb-summary.json
195
+ 2024-03-07 03:39:23,501 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: partial_history
196
+ 2024-03-07 03:39:23,503 DEBUG SenderThread:147 [sender.py:send():382] send: history
197
+ 2024-03-07 03:39:23,503 DEBUG SenderThread:147 [sender.py:send_request():409] send_request: summary_record
198
+ 2024-03-07 03:39:23,504 INFO SenderThread:147 [sender.py:_save_file():1403] saving file wandb-summary.json with policy end
199
+ 2024-03-07 03:39:23,514 INFO Thread-12 :147 [dir_watcher.py:_on_file_modified():288] file/dir modified: /kaggle/working/wandb/run-20240307_033604-u9myy8y3/files/wandb-summary.json
200
+ 2024-03-07 03:39:24,514 INFO Thread-12 :147 [dir_watcher.py:_on_file_modified():288] file/dir modified: /kaggle/working/wandb/run-20240307_033604-u9myy8y3/files/output.log
201
+ 2024-03-07 03:39:25,505 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
202
+ 2024-03-07 03:39:26,475 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
203
+ 2024-03-07 03:39:28,436 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: partial_history
204
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
205
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
206
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
207
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
208
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
209
+ 2024-03-07 03:39:28,439 DEBUG SenderThread:147 [sender.py:send():382] send: history
210
+ 2024-03-07 03:39:28,439 DEBUG SenderThread:147 [sender.py:send_request():409] send_request: summary_record
211
+ 2024-03-07 03:39:28,441 INFO SenderThread:147 [sender.py:_save_file():1403] saving file wandb-summary.json with policy end
212
+ 2024-03-07 03:39:28,515 INFO Thread-12 :147 [dir_watcher.py:_on_file_modified():288] file/dir modified: /kaggle/working/wandb/run-20240307_033604-u9myy8y3/files/wandb-summary.json
213
+ 2024-03-07 03:39:31,442 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
214
+ 2024-03-07 03:39:31,476 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
215
+ 2024-03-07 03:39:35,519 DEBUG SenderThread:147 [sender.py:send():382] send: stats
216
+ 2024-03-07 03:39:36,477 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
217
+ 2024-03-07 03:39:36,520 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
218
+ 2024-03-07 03:39:41,478 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
219
+ 2024-03-07 03:39:41,526 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
wandb/run-20240307_033604-u9myy8y3/files/config.yaml CHANGED
@@ -102,6 +102,26 @@ _wandb:
102
  5: 1
103
  6:
104
  - 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
105
  return_dict:
106
  desc: null
107
  value: true
 
102
  5: 1
103
  6:
104
  - 1
105
+ - 1: train/train_runtime
106
+ 5: 1
107
+ 6:
108
+ - 1
109
+ - 1: train/train_samples_per_second
110
+ 5: 1
111
+ 6:
112
+ - 1
113
+ - 1: train/train_steps_per_second
114
+ 5: 1
115
+ 6:
116
+ - 1
117
+ - 1: train/total_flos
118
+ 5: 1
119
+ 6:
120
+ - 1
121
+ - 1: train/train_loss
122
+ 5: 1
123
+ 6:
124
+ - 1
125
  return_dict:
126
  desc: null
127
  value: true
wandb/run-20240307_033604-u9myy8y3/files/output.log CHANGED
@@ -4,4 +4,6 @@
4
  _warn_prf(average, modifier, msg_start, len(result))
5
  /opt/conda/lib/python3.10/site-packages/sklearn/metrics/_classification.py:1344: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
6
  _warn_prf(average, modifier, msg_start, len(result))
 
 
7
  /opt/conda/lib/python3.10/site-packages/sklearn/metrics/_classification.py:1344: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
 
4
  _warn_prf(average, modifier, msg_start, len(result))
5
  /opt/conda/lib/python3.10/site-packages/sklearn/metrics/_classification.py:1344: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
6
  _warn_prf(average, modifier, msg_start, len(result))
7
+ /opt/conda/lib/python3.10/site-packages/sklearn/metrics/_classification.py:1344: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
8
+ _warn_prf(average, modifier, msg_start, len(result))
9
  /opt/conda/lib/python3.10/site-packages/sklearn/metrics/_classification.py:1344: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
wandb/run-20240307_033604-u9myy8y3/files/wandb-summary.json CHANGED
@@ -1 +1 @@
1
- {"train/loss": 1.5588, "train/grad_norm": 11.676558494567871, "train/learning_rate": 4.064516129032259e-06, "train/epoch": 4.0, "train/global_step": 248, "_timestamp": 1709782727.9843094, "_runtime": 163.96700239181519, "_step": 7, "eval/loss": 1.772017240524292, "eval/accuracy": 0.38181818181818183, "eval/precision": 0.25703185703185705, "eval/recall": 0.38181818181818183, "eval/f1": 0.29505234159779614, "eval/runtime": 1.1221, "eval/samples_per_second": 49.017, "eval/steps_per_second": 6.238}
 
1
+ {"train/loss": 1.5273, "train/grad_norm": 13.581315994262695, "train/learning_rate": 1.2903225806451614e-07, "train/epoch": 5.0, "train/global_step": 310, "_timestamp": 1709782768.4354718, "_runtime": 204.41816473007202, "_step": 10, "eval/loss": 1.7777334451675415, "eval/accuracy": 0.36363636363636365, "eval/precision": 0.246951871657754, "eval/recall": 0.36363636363636365, "eval/f1": 0.2844291976367448, "eval/runtime": 1.1087, "eval/samples_per_second": 49.61, "eval/steps_per_second": 6.314, "train/train_runtime": 220.2648, "train/train_samples_per_second": 11.146, "train/train_steps_per_second": 1.407, "train/total_flos": 645966638976000.0, "train/train_loss": 1.648285896547379}
wandb/run-20240307_033604-u9myy8y3/logs/debug-internal.log CHANGED
@@ -181,3 +181,39 @@
181
  2024-03-07 03:39:03,990 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
182
  2024-03-07 03:39:05,518 DEBUG SenderThread:147 [sender.py:send():382] send: stats
183
  2024-03-07 03:39:06,291 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
181
  2024-03-07 03:39:03,990 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
182
  2024-03-07 03:39:05,518 DEBUG SenderThread:147 [sender.py:send():382] send: stats
183
  2024-03-07 03:39:06,291 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
184
+ 2024-03-07 03:39:09,519 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
185
+ 2024-03-07 03:39:11,292 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
186
+ 2024-03-07 03:39:14,520 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
187
+ 2024-03-07 03:39:16,294 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
188
+ 2024-03-07 03:39:19,521 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
189
+ 2024-03-07 03:39:21,295 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
190
+ 2024-03-07 03:39:22,389 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: partial_history
191
+ 2024-03-07 03:39:22,390 DEBUG SenderThread:147 [sender.py:send():382] send: history
192
+ 2024-03-07 03:39:22,390 DEBUG SenderThread:147 [sender.py:send_request():409] send_request: summary_record
193
+ 2024-03-07 03:39:22,391 INFO SenderThread:147 [sender.py:_save_file():1403] saving file wandb-summary.json with policy end
194
+ 2024-03-07 03:39:22,513 INFO Thread-12 :147 [dir_watcher.py:_on_file_modified():288] file/dir modified: /kaggle/working/wandb/run-20240307_033604-u9myy8y3/files/wandb-summary.json
195
+ 2024-03-07 03:39:23,501 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: partial_history
196
+ 2024-03-07 03:39:23,503 DEBUG SenderThread:147 [sender.py:send():382] send: history
197
+ 2024-03-07 03:39:23,503 DEBUG SenderThread:147 [sender.py:send_request():409] send_request: summary_record
198
+ 2024-03-07 03:39:23,504 INFO SenderThread:147 [sender.py:_save_file():1403] saving file wandb-summary.json with policy end
199
+ 2024-03-07 03:39:23,514 INFO Thread-12 :147 [dir_watcher.py:_on_file_modified():288] file/dir modified: /kaggle/working/wandb/run-20240307_033604-u9myy8y3/files/wandb-summary.json
200
+ 2024-03-07 03:39:24,514 INFO Thread-12 :147 [dir_watcher.py:_on_file_modified():288] file/dir modified: /kaggle/working/wandb/run-20240307_033604-u9myy8y3/files/output.log
201
+ 2024-03-07 03:39:25,505 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
202
+ 2024-03-07 03:39:26,475 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
203
+ 2024-03-07 03:39:28,436 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: partial_history
204
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
205
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
206
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
207
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
208
+ 2024-03-07 03:39:28,438 DEBUG SenderThread:147 [sender.py:send():382] send: metric
209
+ 2024-03-07 03:39:28,439 DEBUG SenderThread:147 [sender.py:send():382] send: history
210
+ 2024-03-07 03:39:28,439 DEBUG SenderThread:147 [sender.py:send_request():409] send_request: summary_record
211
+ 2024-03-07 03:39:28,441 INFO SenderThread:147 [sender.py:_save_file():1403] saving file wandb-summary.json with policy end
212
+ 2024-03-07 03:39:28,515 INFO Thread-12 :147 [dir_watcher.py:_on_file_modified():288] file/dir modified: /kaggle/working/wandb/run-20240307_033604-u9myy8y3/files/wandb-summary.json
213
+ 2024-03-07 03:39:31,442 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
214
+ 2024-03-07 03:39:31,476 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
215
+ 2024-03-07 03:39:35,519 DEBUG SenderThread:147 [sender.py:send():382] send: stats
216
+ 2024-03-07 03:39:36,477 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
217
+ 2024-03-07 03:39:36,520 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
218
+ 2024-03-07 03:39:41,478 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: keepalive
219
+ 2024-03-07 03:39:41,526 DEBUG HandlerThread:147 [handler.py:handle_request():146] handle_request: status_report
wandb/run-20240307_033604-u9myy8y3/run-u9myy8y3.wandb CHANGED
Binary files a/wandb/run-20240307_033604-u9myy8y3/run-u9myy8y3.wandb and b/wandb/run-20240307_033604-u9myy8y3/run-u9myy8y3.wandb differ