Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,33 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
license: bsd-2-clause
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
tags:
|
| 5 |
+
- pytorch
|
| 6 |
+
- text-generation
|
| 7 |
+
- causal-lm
|
| 8 |
+
- rwkv
|
| 9 |
license: bsd-2-clause
|
| 10 |
+
datasets:
|
| 11 |
+
- The Pile
|
| 12 |
+
|
| 13 |
---
|
| 14 |
+
|
| 15 |
+
# RWKV-4 430M
|
| 16 |
+
|
| 17 |
+
## Model Description
|
| 18 |
+
|
| 19 |
+
RWKV-4 430M is a L24-D1024 causal language model trained on the Pile. See https://github.com/BlinkDL/RWKV-LM for details.
|
| 20 |
+
|
| 21 |
+
At this moment you have to use my Github code (https://github.com/BlinkDL/RWKV-LM) to run it.
|
| 22 |
+
|
| 23 |
+
ctx_len = 1024
|
| 24 |
+
n_layer = 24
|
| 25 |
+
n_embd = 1024
|
| 26 |
+
|
| 27 |
+
Final checkpoint:
|
| 28 |
+
RWKV-4-Pile-430M-20220808-8066.pth : Trained on the Pile for 333B tokens.
|
| 29 |
+
* Pile loss 2.2621
|
| 30 |
+
* LAMBADA ppl 13.04, acc 45.16%
|
| 31 |
+
* PIQA acc 67.52%
|
| 32 |
+
* SC2016 acc 63.87%
|
| 33 |
+
* Hellaswag acc_norm 40.90%
|