Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: transformers
|
| 3 |
+
tags: []
|
| 4 |
+
---
|
| 5 |
+
# FW Medium
|
| 6 |
+
|
| 7 |
+
This is the tiny version of the bilinear transformers trained on FineWeb-edu.
|
| 8 |
+
The primary purpose of this model is interpretability, most design choices were made with that in mind.
|
| 9 |
+
|
| 10 |
+
The code to run this custom model can be found [here](https://github.com/tdooms/bilinear-decomposition), along with many utility functions for weight-based interpretability.
|
| 11 |
+
|
| 12 |
+
## Model Details
|
| 13 |
+
- 335 million parameters
|
| 14 |
+
- 16 layers
|
| 15 |
+
- 16 attention heads
|
| 16 |
+
- model dimension 1024
|
| 17 |
+
- bilinear MLP with expansion factor 4
|
| 18 |
+
- context length of 512
|
| 19 |
+
- trained for 132B tokens
|
| 20 |
+
- rotary positional embedding
|
| 21 |
+
- Mixtral [tokenizer](https://huggingface.co/mistral-community/Mixtral-8x22B-v0.1)
|