rwkv-4-pile-7b / README.md
BlinkDL's picture
Update README.md
26ef629
|
raw
history blame
739 Bytes
metadata
language:
  - en
tags:
  - pytorch
  - text-generation
  - causal-lm
  - rwkv
license: apache-2.0
datasets:
  - The Pile

RWKV-4 7B

Model Description

RWKV-4 7B is a L32-D4096 causal language model trained on the Pile. See https://github.com/BlinkDL/RWKV-LM for details.

** Note: It's a BF16 model, and it may overflow if you are using FP16 (probably fixable by rescaling the weights). **

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

ctx_len = 1024 n_layer = 32 n_embd = 4096

Preview checkpoint: RWKV-4-Pile-7B-20220928-2251.pth : Trained on the Pile for 93B tokens.

  • Pile loss 1.9432
  • LAMBADA ppl 5.05, acc 63.73%
  • PIQA acc 74.10%
  • SC2016 acc 71.62%
  • Hellaswag acc_norm 60.71%