Toflamus commited on
Commit
b768af7
·
1 Parent(s): 5365012

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +42 -2
README.md CHANGED
@@ -14,7 +14,7 @@ should probably proofread and complete it, then remove this comment. -->
14
  # Output
15
 
16
  This model is a fine-tuned version of [Toflamus/GPT-2_para3M](https://huggingface.co/Toflamus/GPT-2_para3M) on an unknown dataset.
17
-
18
  ## Model description
19
 
20
  More information needed
@@ -44,7 +44,47 @@ The following hyperparameters were used during training:
44
  - num_epochs: 5
45
 
46
  ### Training results
47
-
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
48
 
49
 
50
  ### Framework versions
 
14
  # Output
15
 
16
  This model is a fine-tuned version of [Toflamus/GPT-2_para3M](https://huggingface.co/Toflamus/GPT-2_para3M) on an unknown dataset.
17
+ TrainOutput(global_step=4060, training_loss=6.123095868491187, metrics={'train_runtime': 1435.0504, 'train_samples_per_second': 181.185, 'train_steps_per_second': 2.829, 'total_flos': 96669633527808.0, 'train_loss': 6.123095868491187, 'epoch': 5.0})
18
  ## Model description
19
 
20
  More information needed
 
44
  - num_epochs: 5
45
 
46
  ### Training results
47
+ Step Training Loss
48
+ 100 7.737900
49
+ 200 7.066700
50
+ 300 6.840200
51
+ 400 6.686600
52
+ 500 6.607700
53
+ 600 6.516500
54
+ 700 6.449800
55
+ 800 6.360400
56
+ 900 6.321700
57
+ 1000 6.252700
58
+ 1100 6.223500
59
+ 1200 6.194700
60
+ 1300 6.131500
61
+ 1400 6.113400
62
+ 1500 6.106500
63
+ 1600 6.044100
64
+ 1700 6.024400
65
+ 1800 6.008500
66
+ 1900 6.006600
67
+ 2000 5.959900
68
+ 2100 5.931100
69
+ 2200 5.925300
70
+ 2300 5.933500
71
+ 2400 5.921900
72
+ 2500 5.913400
73
+ 2600 5.898100
74
+ 2700 5.874700
75
+ 2800 5.869100
76
+ 2900 5.851200
77
+ 3000 5.853900
78
+ 3100 5.870100
79
+ 3200 5.868100
80
+ 3300 5.837000
81
+ 3400 5.845300
82
+ 3500 5.828800
83
+ 3600 5.847400
84
+ 3700 5.858600
85
+ 3800 5.853200
86
+ 3900 5.836600
87
+ 4000 5.849100
88
 
89
 
90
  ### Framework versions