Hugging Face's logo

BlinkDL
/
rwkv-4-pile-3b

Text Generation
PyTorch
English
causal-lm
rwkv
Model card Files Community
2
BlinkDL commited on
Commit
bd0d392
·
1 Parent(s): 6fe5107

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +2 -2
README.md CHANGED
@@ -27,7 +27,7 @@ New checkpoint: RWKV-4-Pile-3B-20221110-ctx4096.pth : Fine-tuned to ctx_len = 40
27
  * PIQA acc 74.16%
28
  * SC2016 acc 70.71%
29
  * Hellaswag acc_norm 59.89%
30
- ctx_len = 4096 n_layer = 32 n_embd = 2560
31
 
32
  Final checkpoint: RWKV-4-Pile-3B-20221008-8023.pth : Trained on the Pile for 331B tokens.
33
  * Pile loss 1.9469
@@ -35,4 +35,4 @@ Final checkpoint: RWKV-4-Pile-3B-20221008-8023.pth : Trained on the Pile for 331
35
  * PIQA acc 73.72%
36
  * SC2016 acc 70.28%
37
  * Hellaswag acc_norm 59.63%
38
- ctx_len = 1024 n_layer = 32 n_embd = 2560
 
27
  * PIQA acc 74.16%
28
  * SC2016 acc 70.71%
29
  * Hellaswag acc_norm 59.89%
30
+ * ctx_len = 4096 n_layer = 32 n_embd = 2560
31
 
32
  Final checkpoint: RWKV-4-Pile-3B-20221008-8023.pth : Trained on the Pile for 331B tokens.
33
  * Pile loss 1.9469
 
35
  * PIQA acc 73.72%
36
  * SC2016 acc 70.28%
37
  * Hellaswag acc_norm 59.63%
38
+ * ctx_len = 1024 n_layer = 32 n_embd = 2560