Update README.md
Browse files
README.md
CHANGED
|
@@ -25,3 +25,10 @@ At this moment you have to use my Github code (https://github.com/BlinkDL/RWKV-L
|
|
| 25 |
ctx_len = 1024
|
| 26 |
n_layer = 40
|
| 27 |
n_embd = 5120
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
ctx_len = 1024
|
| 26 |
n_layer = 40
|
| 27 |
n_embd = 5120
|
| 28 |
+
|
| 29 |
+
Preview checkpoint: RWKV-4-Pile-14B-20221103-1214.pth : Trained on the Pile for 50B tokens.
|
| 30 |
+
* Pile loss 1.9533
|
| 31 |
+
* LAMBADA ppl 5.07, acc 64.08%
|
| 32 |
+
* PIQA acc 73.99%
|
| 33 |
+
* SC2016 acc 72.05%
|
| 34 |
+
* Hellaswag acc_norm 61.07%
|