Downloads · 30 days
11
8% of all-time downloads
rwkv-3-pile-430m
BlinkDL/rwkv-3-pile-430m
rwkv-3-pile-430m is a text generation model from BlinkDL. Use it when you need the model to write or continue text. It is set up for rwkv. The card lists the license as apache-2.0.
RWKV-3 430M is a L24-D1024 causal language model trained on the Pile. See https://github.com/BlinkDL/RWKV-LM for details.
All-time downloads
146
Public
Repo size
3.4 GB
Likes
4
Public
Repo makeup
Click a slice to open those files.
.pth3.4 GB · 100%
At a glance
- Task
- Text Generation
- Library
- rwkv
- License
- apache-2.0
- Access
- Public
- Created
- Jul 20, 2022
- Updated
- Sep 14, 2022
- SHA
- cb7117a3
Try a prompt
Model card
From the Hugging Face model README
RWKV-3 430M
Model Description
RWKV-3 430M is a L24-D1024 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 = 24 n_embd = 1024
Final checkpoint: RWKV-3-Pile-430M-20220817-10602.pth : Trained on the Pile for 326B tokens.
- Pile loss 2.288
- LAMBADA ppl 12.83, acc 45.45%
- PIQA acc 68.50%
- SC2016 acc 63.60%
- Hellaswag acc_norm 40.68%
Preview checkpoint: RWKV-3-Pile-20220721-3029.pth : Trained on the Pile for 93B tokens.
- Pile loss 2.341
- LAMBADA ppl 14.18, acc 44.25%
- PIQA acc 67.95%
- SC2016 acc 63.39%
- Hellaswag acc_norm 39.06%
- Task
- Text Generation
- Library
- rwkv
- License
- apache-2.0
- Languages
- en
- Created
- Jul 20, 2022
- Updated
- Sep 14, 2022