Hugging Face's logo

augustocsc
/
gpt2-p10k-cossine

Text Generation
Transformers
Safetensors
gpt2
Generated from Trainer
text-generation-inference
Inference Endpoints
Model card Files Community
augustocsc commited on
Commit
a57f8dc
•
1 Parent(s): 3f48ec7

End of training

Browse files
Files changed (1) hide show
  1. README.md +57 -13
README.md CHANGED
@@ -15,7 +15,7 @@ should probably proofread and complete it, then remove this comment. -->
15
 
16
  This model is a fine-tuned version of [gpt2](https://huggingface.co/gpt2) on an unknown dataset.
17
  It achieves the following results on the evaluation set:
18
- - Loss: 14.1032
19
 
20
  ## Model description
21
 
@@ -35,23 +35,67 @@ More information needed
35
 
36
  The following hyperparameters were used during training:
37
  - learning_rate: 5e-05
38
- - train_batch_size: 256
39
- - eval_batch_size: 256
40
  - seed: 42
41
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
42
- - lr_scheduler_type: constant_with_warmup
43
- - num_epochs: 5
44
 
45
  ### Training results
46
 
47
- | Training Loss | Epoch | Step | Validation Loss |
48
- |:-------------:|:-----:|:----:|:---------------:|
49
- | No log | 0.8 | 200 | 0.0382 |
50
- | No log | 1.6 | 400 | 0.0851 |
51
- | 0.4686 | 2.4 | 600 | 1.0672 |
52
- | 0.4686 | 3.2 | 800 | 2.6885 |
53
- | 7.319 | 4.0 | 1000 | 2.5899 |
54
- | 7.319 | 4.8 | 1200 | 14.1032 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
55
 
56
 
57
  ### Framework versions
 
15
 
16
  This model is a fine-tuned version of [gpt2](https://huggingface.co/gpt2) on an unknown dataset.
17
  It achieves the following results on the evaluation set:
18
+ - Loss: 0.0234
19
 
20
  ## Model description
21
 
 
35
 
36
  The following hyperparameters were used during training:
37
  - learning_rate: 5e-05
38
+ - train_batch_size: 64
39
+ - eval_batch_size: 64
40
  - seed: 42
41
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
42
+ - lr_scheduler_type: cosine
43
+ - num_epochs: 10
44
 
45
  ### Training results
46
 
47
+ | Training Loss | Epoch | Step | Validation Loss |
48
+ |:-------------:|:-----:|:-----:|:---------------:|
49
+ | No log | 0.2 | 200 | 0.0386 |
50
+ | No log | 0.4 | 400 | 0.0523 |
51
+ | 0.1425 | 0.6 | 600 | 1.0542 |
52
+ | 0.1425 | 0.8 | 800 | 1.0459 |
53
+ | 0.9177 | 1.0 | 1000 | 0.3677 |
54
+ | 0.9177 | 1.2 | 1200 | 0.0296 |
55
+ | 0.9177 | 1.4 | 1400 | 0.0284 |
56
+ | 0.0421 | 1.6 | 1600 | 0.0275 |
57
+ | 0.0421 | 1.8 | 1800 | 0.0274 |
58
+ | 0.029 | 2.0 | 2000 | 0.0264 |
59
+ | 0.029 | 2.2 | 2200 | 0.0258 |
60
+ | 0.029 | 2.4 | 2400 | 0.0256 |
61
+ | 0.0276 | 2.6 | 2600 | 0.0254 |
62
+ | 0.0276 | 2.8 | 2800 | 0.0252 |
63
+ | 0.0265 | 3.0 | 3000 | 0.0251 |
64
+ | 0.0265 | 3.2 | 3200 | 0.0247 |
65
+ | 0.0265 | 3.4 | 3400 | 0.0247 |
66
+ | 0.0256 | 3.6 | 3600 | 0.0246 |
67
+ | 0.0256 | 3.8 | 3800 | 0.0252 |
68
+ | 0.0262 | 4.0 | 4000 | 0.0249 |
69
+ | 0.0262 | 4.2 | 4200 | 0.0244 |
70
+ | 0.0262 | 4.4 | 4400 | 0.0242 |
71
+ | 0.0255 | 4.6 | 4600 | 0.0242 |
72
+ | 0.0255 | 4.8 | 4800 | 0.0248 |
73
+ | 0.0251 | 5.0 | 5000 | 0.0240 |
74
+ | 0.0251 | 5.2 | 5200 | 0.0242 |
75
+ | 0.0251 | 5.4 | 5400 | 0.0239 |
76
+ | 0.0254 | 5.6 | 5600 | 0.0239 |
77
+ | 0.0254 | 5.8 | 5800 | 0.0237 |
78
+ | 0.0244 | 6.0 | 6000 | 0.0237 |
79
+ | 0.0244 | 6.2 | 6200 | 0.0239 |
80
+ | 0.0244 | 6.4 | 6400 | 0.0237 |
81
+ | 0.0244 | 6.6 | 6600 | 0.0237 |
82
+ | 0.0244 | 6.8 | 6800 | 0.0238 |
83
+ | 0.0246 | 7.0 | 7000 | 0.0236 |
84
+ | 0.0246 | 7.2 | 7200 | 0.0235 |
85
+ | 0.0246 | 7.4 | 7400 | 0.0235 |
86
+ | 0.0242 | 7.6 | 7600 | 0.0235 |
87
+ | 0.0242 | 7.8 | 7800 | 0.0235 |
88
+ | 0.0244 | 8.0 | 8000 | 0.0236 |
89
+ | 0.0244 | 8.2 | 8200 | 0.0234 |
90
+ | 0.0244 | 8.4 | 8400 | 0.0235 |
91
+ | 0.0246 | 8.6 | 8600 | 0.0234 |
92
+ | 0.0246 | 8.8 | 8800 | 0.0234 |
93
+ | 0.0237 | 9.0 | 9000 | 0.0234 |
94
+ | 0.0237 | 9.2 | 9200 | 0.0234 |
95
+ | 0.0237 | 9.4 | 9400 | 0.0234 |
96
+ | 0.0241 | 9.6 | 9600 | 0.0234 |
97
+ | 0.0241 | 9.8 | 9800 | 0.0234 |
98
+ | 0.024 | 10.0 | 10000 | 0.0234 |
99
 
100
 
101
  ### Framework versions