OpenSoftware-World commited on
Commit
6786201
·
verified ·
1 Parent(s): 7261976

Create model_specifications.txt

Browse files
models/opensoftware-world-osw1:5m/model_specifications.txt ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ block_size = 256
2
+ d_model = 256
3
+ n_layer = 4
4
+ n_head = 4
5
+ d_ff = 1024
6
+ dropout = 0.1
7
+ batch_size = 16
8
+ grad_accum_steps = 2
9
+ epochs = 40
10
+ max_lr = 2e-4
11
+ min_lr = 2e-5
12
+ warmup_ratio = 0.05
13
+ weight_decay = 0.05
14
+ grad_clip = 1.0
15
+ label_smoothing = 0.0