Hugging Face's logo

BlinkDL
/
rwkv-4-pile-14b

Text Generation
PyTorch
English
causal-lm
rwkv
Model card Files Community
2
BlinkDL commited on
Commit
f720a95
•
1 Parent(s): 807a65e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +2 -1
README.md CHANGED
@@ -20,14 +20,15 @@ RWKV-4 14B is a L40-D5120 causal language model trained on the Pile. See https:/
20
 
21
  Use https://github.com/BlinkDL/ChatRWKV to run it.
22
 
23
- ctx_len = 1024
24
  n_layer = 40
25
  n_embd = 5120
26
 
27
  RWKV-4-Pile-14B-2023xxxx-ctx4096-testxxx.pth : Fine-tuned to ctx_len 4096.
 
28
  * Highly recommended. It's great.
29
 
30
  RWKV-4-Pile-14B-20230213-8019.pth : Trained on the Pile for 331B tokens.
 
31
  * Pile loss 1.7579
32
  * LAMBADA ppl 3.81, acc 71.05%
33
  * PIQA acc 77.42%
 
20
 
21
  Use https://github.com/BlinkDL/ChatRWKV to run it.
22
 
 
23
  n_layer = 40
24
  n_embd = 5120
25
 
26
  RWKV-4-Pile-14B-2023xxxx-ctx4096-testxxx.pth : Fine-tuned to ctx_len 4096.
27
+ * ctx_len = 4096
28
  * Highly recommended. It's great.
29
 
30
  RWKV-4-Pile-14B-20230213-8019.pth : Trained on the Pile for 331B tokens.
31
+ * ctx_len = 1024
32
  * Pile loss 1.7579
33
  * LAMBADA ppl 3.81, acc 71.05%
34
  * PIQA acc 77.42%