Update README.md

README.md

@@ -24,6 +24,15 @@ ctx_len = 896
 n_layer = 24
 n_embd = 2048
 
-20220708-1905.pth : Trained on the Pile for 68B tokens. Pile loss 2.148, LAMBADA ppl 8.41, acc 53.17%.
+Preview checkpoint: RWKV-3-Pile-20220723-3542.pth : Trained on the Pile for 127B tokens.
+* Pile loss 2.102 
+* LAMBADA ppl 7.52, acc 54.71%
+* PIQA acc 71.11%
+* SC2016 acc 67.24%
+* Hellaswag acc_norm 50.45%
+
+Preview checkpoint: 20220708-1905.pth : Trained on the Pile for 68B tokens.
+* Pile loss 2.148
+* LAMBADA ppl 8.41, acc 53.17%
 
 (I am still training it)