Link the post
Browse files
README.md
CHANGED
|
@@ -11,7 +11,7 @@ tags:
|
|
| 11 |
|
| 12 |
Eight byte-level H-Nets ([Hwang, Wang, Gu](https://arxiv.org/abs/2507.07955)) trained on SMILES. All share one two-stage architecture (bytes to pieces to groups, Transformer stacks, cosine router) and one dataset. They differ in the training objective, and so in where they cut a molecule.
|
| 13 |
|
| 14 |
-
Code: [DrHB/hnet-smiles](https://github.com/DrHB/hnet-smiles). Post: How the training objective shapes learned tokens
|
| 15 |
|
| 16 |
## Models
|
| 17 |
|
|
|
|
| 11 |
|
| 12 |
Eight byte-level H-Nets ([Hwang, Wang, Gu](https://arxiv.org/abs/2507.07955)) trained on SMILES. All share one two-stage architecture (bytes to pieces to groups, Transformer stacks, cosine router) and one dataset. They differ in the training objective, and so in where they cut a molecule.
|
| 13 |
|
| 14 |
+
Code: [DrHB/hnet-smiles](https://github.com/DrHB/hnet-smiles). Post: [How the training objective shapes learned tokens](https://drhb.github.io/blogposts/posts/learned-cuts/).
|
| 15 |
|
| 16 |
## Models
|
| 17 |
|