trollek commited on
Commit
b689a9c
·
verified ·
1 Parent(s): c4d024d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +31 -1
README.md CHANGED
@@ -72,4 +72,34 @@ val_size: 0.01
72
  per_device_eval_batch_size: 1
73
  eval_strategy: steps
74
  eval_steps: 1000
75
- ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
72
  per_device_eval_batch_size: 1
73
  eval_strategy: steps
74
  eval_steps: 1000
75
+ ```
76
+
77
+ ### BAdam training results
78
+
79
+ | Training Loss | Epoch | Step | Validation Loss |
80
+ |:-------------:|:------:|:-----:|:---------------:|
81
+ | 0.2748 | 0.0617 | 1000 | 0.2788 |
82
+ | 0.2786 | 0.1234 | 2000 | 0.2503 |
83
+ | 0.18 | 0.1850 | 3000 | 0.2144 |
84
+ | 0.2015 | 0.2467 | 4000 | 0.1926 |
85
+ | 0.2044 | 0.3084 | 5000 | 0.1777 |
86
+ | 0.142 | 0.3701 | 6000 | 0.1661 |
87
+ | 0.1813 | 0.4317 | 7000 | 0.1570 |
88
+ | 0.1413 | 0.4934 | 8000 | 0.1529 |
89
+ | 0.1805 | 0.5551 | 9000 | 0.1462 |
90
+ | 0.1431 | 0.6168 | 10000 | 0.1410 |
91
+ | 0.1693 | 0.6784 | 11000 | 0.1375 |
92
+ | 0.1291 | 0.7401 | 12000 | 0.1357 |
93
+ | 0.1501 | 0.8018 | 13000 | 0.1348 |
94
+ | 0.1521 | 0.8635 | 14000 | 0.1345 |
95
+ | 0.1279 | 0.9251 | 15000 | 0.1346 |
96
+ | 0.1351 | 0.9868 | 16000 | 0.1344 |
97
+
98
+ ### GSM8K results
99
+
100
+ |Tasks|Version| Filter |n-shot| Metric |Value | |Stderr|
101
+ |-----|------:|----------------|-----:|-----------|-----:|---|-----:|
102
+ |gsm8k| 3|strict-match | 5|exact_match|0.2691|± |0.0122|
103
+ | | |flexible-extract| 5|exact_match|0.2752|± |0.0123|
104
+
105
+ It matches the chat trained model from h2o.