RWKV-3 169M

Model Description

RWKV-3 169M is a L12-D768 causal language model trained on the Pile. See https://github.com/BlinkDL/RWKV-LM for details.

At this moment you have to use my Github code (https://github.com/BlinkDL/RWKV-v2-RNN-Pile) to run it.

ctx_len = 768 n_layer = 12 n_embd = 768

Final checkpoint: RWKV-3-Pile-20220720-10704.pth : Trained on the Pile for 328B tokens.

• Pile loss 2.5596
• LAMBADA ppl 28.82, acc 32.33%
• PIQA acc 64.15%
• SC2016 acc 57.88%
• Hellaswag acc_norm 32.45%

Preview checkpoint: 20220703-1652.pth : Trained on the Pile for 50B tokens. Pile loss 2.6375, LAMBADA ppl 33.30, acc 31.24%.