Edit model card
YAML Metadata Error: "datasets[0]" with value "The Pile" is not valid. If possible, use a dataset id from https://hf.co/datasets.

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%.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Examples
Unable to determine this model's library. Check the docs .

Spaces using BlinkDL/rwkv-3-pile-169m 2