xuancoblab2023 commited on
Commit
c2ff74a
·
verified ·
1 Parent(s): 395b0dc

Training in progress, epoch 2

Browse files
logs/events.out.tfevents.1709872398.6c1c016b0e3e.7263.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:aba27c272f71ae8cd6972aa68b8f1b09c93b6c05245499b5a550cad17221f3a2
3
- size 5314
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1915b6a2c06d155db3570a3e7513c6285734d3e6cfb61979d5cb5ebe64b0f2a2
3
+ size 5984
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:45b4b95ef18e9ad28c7043735a6cae19edb8de8cca02e86dd2739757a4f1dc6d
3
  size 17549312
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a9deaf7b9483814e8658af2bd540c9c428e1d9d74902b8546da37c62a418f2b8
3
  size 17549312
run-0/checkpoint-96/model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c92adc7dcf01e1fbb05511c4493199b42d344e3981283ab521c477e538f9aceb
3
  size 17549312
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a9deaf7b9483814e8658af2bd540c9c428e1d9d74902b8546da37c62a418f2b8
3
  size 17549312
run-0/checkpoint-96/optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:db3824b1c6f237bcf65a54ea4cca3aa03e4cfa3c81a3b6afc4e72e03c759249e
3
  size 35122746
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5e8f5ef6ee86d49f7eb75c8214eefcb88b1a2166ac8c9a098a261f23125fce34
3
  size 35122746
run-0/checkpoint-96/rng_state.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:24aa86019b8aea1c551cc1adaf38c4db2fc01de75a22af312230f6b592e0fd81
3
  size 14054
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6bc13e09b59a62b6fb29cad756bfeefcce6f512c71a9054cc39e336db0b532a5
3
  size 14054
run-0/checkpoint-96/scheduler.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:43ddd3b1c41613cd7e1e1227d929645b76a450f62d1a6ce5575c2f7873d9d0b6
3
  size 1064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5bd902627d0c1d565c70987837a95a63c8ee21875132d3259ea0ce69d0e7372a
3
  size 1064
run-0/checkpoint-96/trainer_state.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
- "best_metric": 0.7514677103718199,
3
- "best_model_checkpoint": "tiny-bert-sst2-distilled/run-0/checkpoint-96",
4
- "epoch": 1.0,
5
  "eval_steps": 500,
6
  "global_step": 96,
7
  "is_hyper_param_search": true,
@@ -10,36 +10,55 @@
10
  "log_history": [
11
  {
12
  "epoch": 1.0,
13
- "grad_norm": 1.6740443706512451,
14
- "learning_rate": 0.0001218354408608861,
15
- "loss": 0.4816,
16
- "step": 96
17
  },
18
  {
19
  "epoch": 1.0,
20
- "eval_accuracy": 0.7514677103718199,
21
- "eval_f1": 0.7519531250000001,
22
- "eval_loss": 0.41538161039352417,
23
- "eval_precision": 0.7504873294346979,
24
- "eval_recall": 0.7534246575342466,
25
- "eval_runtime": 26.5175,
26
- "eval_samples_per_second": 38.541,
27
- "eval_steps_per_second": 1.207,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28
  "step": 96
29
  }
30
  ],
31
  "logging_steps": 500,
32
- "max_steps": 480,
33
  "num_input_tokens_seen": 0,
34
- "num_train_epochs": 5,
35
  "save_steps": 500,
36
- "total_flos": 235695197280.0,
37
- "train_batch_size": 32,
38
  "trial_name": null,
39
  "trial_params": {
40
- "alpha": 0.679174768290245,
41
- "learning_rate": 0.00015229430107610762,
42
- "num_train_epochs": 5,
43
- "temperature": 27
44
  }
45
  }
 
1
  {
2
+ "best_metric": 0.5,
3
+ "best_model_checkpoint": "tiny-bert-sst2-distilled/run-0/checkpoint-48",
4
+ "epoch": 2.0,
5
  "eval_steps": 500,
6
  "global_step": 96,
7
  "is_hyper_param_search": true,
 
10
  "log_history": [
11
  {
12
  "epoch": 1.0,
13
+ "grad_norm": 1.7277214527130127,
14
+ "learning_rate": 7.1080634981543736e-06,
15
+ "loss": 0.4421,
16
+ "step": 48
17
  },
18
  {
19
  "epoch": 1.0,
20
+ "eval_accuracy": 0.5,
21
+ "eval_f1": 0.0,
22
+ "eval_loss": 0.4095905125141144,
23
+ "eval_precision": 0.0,
24
+ "eval_recall": 0.0,
25
+ "eval_runtime": 28.1077,
26
+ "eval_samples_per_second": 36.36,
27
+ "eval_steps_per_second": 0.569,
28
+ "step": 48
29
+ },
30
+ {
31
+ "epoch": 2.0,
32
+ "grad_norm": 0.3527555465698242,
33
+ "learning_rate": 3.5540317490771868e-06,
34
+ "loss": 0.4039,
35
+ "step": 96
36
+ },
37
+ {
38
+ "epoch": 2.0,
39
+ "eval_accuracy": 0.5,
40
+ "eval_f1": 0.0,
41
+ "eval_loss": 0.39591261744499207,
42
+ "eval_precision": 0.0,
43
+ "eval_recall": 0.0,
44
+ "eval_runtime": 27.762,
45
+ "eval_samples_per_second": 36.813,
46
+ "eval_steps_per_second": 0.576,
47
  "step": 96
48
  }
49
  ],
50
  "logging_steps": 500,
51
+ "max_steps": 144,
52
  "num_input_tokens_seen": 0,
53
+ "num_train_epochs": 3,
54
  "save_steps": 500,
55
+ "total_flos": 471390394560.0,
56
+ "train_batch_size": 64,
57
  "trial_name": null,
58
  "trial_params": {
59
+ "alpha": 0.48386139243976256,
60
+ "learning_rate": 1.066209524723156e-05,
61
+ "num_train_epochs": 3,
62
+ "temperature": 23
63
  }
64
  }
run-0/checkpoint-96/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1873f3a503b2d5b3f7200baa33f737ba864dd65edb3834d85c5b8e40b6b72f07
3
  size 4920
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:103518af05e87893001984e8fe4e4fbdef270179595accd70434e2248c87999d
3
  size 4920