Update README.md
Browse files
README.md
CHANGED
@@ -24,10 +24,10 @@ ctx_len = 1024
|
|
24 |
n_layer = 40
|
25 |
n_embd = 5120
|
26 |
|
27 |
-
|
28 |
-
* Pile loss 1.
|
29 |
-
* LAMBADA ppl 3.
|
30 |
-
* PIQA acc 77.
|
31 |
-
* SC2016 acc
|
32 |
-
* Hellaswag acc_norm
|
33 |
-
* WinoGrande acc
|
|
|
24 |
n_layer = 40
|
25 |
n_embd = 5120
|
26 |
|
27 |
+
Final checkpoint: RWKV-4-Pile-14B-20230213-8019.pth : Trained on the Pile for 331B tokens.
|
28 |
+
* Pile loss 1.7579
|
29 |
+
* LAMBADA ppl 3.81, acc 71.05%
|
30 |
+
* PIQA acc 77.42%
|
31 |
+
* SC2016 acc 75.57%
|
32 |
+
* Hellaswag acc_norm 70.24%
|
33 |
+
* WinoGrande acc 62.98%
|