jteng commited on
Commit
3e42fd3
·
verified ·
1 Parent(s): df9ee80

Final model after 129534 steps (run records)

Browse files
Files changed (2) hide show
  1. run/progress.json +22 -32
  2. run/result.json +25 -0
run/progress.json CHANGED
@@ -1,35 +1,25 @@
1
  {
2
- "state": "training",
3
- "global_step": 129500,
4
- "max_steps": 129534,
5
- "epoch": 1.9994801852804942,
6
- "elapsed_seconds": 41348.64713783101,
7
- "steps_per_second": 2.406366517103403,
8
- "optimizer_updates_this_process": 99500,
9
- "peak_cuda_GiB": 16.14784526824951,
 
 
 
 
 
 
 
 
 
 
 
10
  "world_size": 1,
11
- "effective_batch_size": 72,
12
- "recent_metrics": [
13
- {
14
- "loss": 0.011637792587280274,
15
- "grad_norm": 0.002693501301109791,
16
- "learning_rate": 3.9140302932048736e-07,
17
- "epoch": 1.9608800823468862,
18
- "step": 127000
19
- },
20
- {
21
- "loss": 0.011701138496398925,
22
- "grad_norm": 0.15038153529167175,
23
- "learning_rate": 2.3700341223153768e-07,
24
- "epoch": 1.9763201235203294,
25
- "step": 128000
26
- },
27
- {
28
- "loss": 0.012043976783752441,
29
- "grad_norm": 0.1061779111623764,
30
- "learning_rate": 8.260379514258804e-08,
31
- "epoch": 1.9917601646937726,
32
- "step": 129000
33
- }
34
- ]
35
  }
 
1
  {
2
+ "scope": "universal",
3
+ "ordinary_version": null,
4
+ "complete": true,
5
+ "smoke_only": false,
6
+ "task": "statement",
7
+ "files": 50000,
8
+ "encoded_rows": 4663180,
9
+ "epoch": 2.0,
10
+ "global_step": 129534,
11
+ "optimizer_updates_this_process": 99534,
12
+ "steps_this_process": 99534,
13
+ "training_loss": 0.010466809748889123,
14
+ "metrics": {
15
+ "train_runtime": 41361.4687,
16
+ "train_samples_per_second": 225.484,
17
+ "train_steps_per_second": 3.132,
18
+ "train_loss": 0.010466809748889123,
19
+ "epoch": 2.0
20
+ },
21
  "world_size": 1,
22
+ "global_batch_size": 72,
23
+ "peak_cuda_GiB": 16.14784526824951,
24
+ "final_weights_sha256": "0e9be71c9c4f255dc691eb59a0894ae9ae56938d0e4a239e2441a2e84ba9dc26"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
25
  }
run/result.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "scope": "universal",
3
+ "ordinary_version": null,
4
+ "complete": true,
5
+ "smoke_only": false,
6
+ "task": "statement",
7
+ "files": 50000,
8
+ "encoded_rows": 4663180,
9
+ "epoch": 2.0,
10
+ "global_step": 129534,
11
+ "optimizer_updates_this_process": 99534,
12
+ "steps_this_process": 99534,
13
+ "training_loss": 0.010466809748889123,
14
+ "metrics": {
15
+ "train_runtime": 41361.4687,
16
+ "train_samples_per_second": 225.484,
17
+ "train_steps_per_second": 3.132,
18
+ "train_loss": 0.010466809748889123,
19
+ "epoch": 2.0
20
+ },
21
+ "world_size": 1,
22
+ "global_batch_size": 72,
23
+ "peak_cuda_GiB": 16.14784526824951,
24
+ "final_weights_sha256": "0e9be71c9c4f255dc691eb59a0894ae9ae56938d0e4a239e2441a2e84ba9dc26"
25
+ }