First commit with tldr of model
Browse files
README.md
CHANGED
|
@@ -1,3 +1,23 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
tags:
|
| 6 |
+
- babylm
|
| 7 |
---
|
| 8 |
+
|
| 9 |
+
# Lil-Bevo
|
| 10 |
+
|
| 11 |
+
Lil-Bevo is UT Austin's submission to the BabyLM challenge, specifically the *strict-small* track.
|
| 12 |
+
|
| 13 |
+
[Link to GitHub Repo](https://github.com/venkatasg/Lil-Bevo)
|
| 14 |
+
|
| 15 |
+
## TLDR:
|
| 16 |
+
- Unigram tokenizer trained on 10M BabyLM tokens plus MAESTRO dataset for a vocab size of 16k.
|
| 17 |
+
- `deberta-small-v3` trained on mixture of MAESTRO and 10M tokens for 3 epochs.
|
| 18 |
+
- Model continues training for 50 epochs on 10M tokens with 128 sequence length.
|
| 19 |
+
- Model continues training for 200 epochs on 10M tokens with 512 sequence length.
|
| 20 |
+
- Model is trained with targeted linguistic masking for 10 epochs.
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
This README will be updated with more details soon.
|