Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,43 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
|
| 5 |
+
# TinyLlama + Japanese
|
| 6 |
+
|
| 7 |
+
A continual pretraining model of TinyLlama 1.1B with a few Japanese texts.
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
### Base Model
|
| 11 |
+
|
| 12 |
+
[TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T](https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T)
|
| 13 |
+
|
| 14 |
+
### Tokenizers
|
| 15 |
+
|
| 16 |
+
(elyza/ELYZA-japanese-Llama-2-7b)[https://huggingface.co/elyza/ELYZA-japanese-Llama-2-7b]
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
### Training Dataset
|
| 20 |
+
|
| 21 |
+
Around 9B tokens in total.
|
| 22 |
+
|
| 23 |
+
- izumi-lab/wikipedia-ja-20230720
|
| 24 |
+
- if001/oscar_2023_filtered
|
| 25 |
+
|
| 26 |
+
### Validation Dataset
|
| 27 |
+
|
| 28 |
+
- izumi-lab/wikinews-ja-20230728
|
| 29 |
+
- izumi-lab/wikinews-en-20230728
|
| 30 |
+
- if001/aozorabunko-clean-sin
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
### Evaluation
|
| 34 |
+
|
| 35 |
+
We did not perform.
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
### Acknowledgement
|
| 39 |
+
|
| 40 |
+
We acknowledge those who prepared valuable datasets and [lit-gpt](https://github.com/Lightning-AI/lit-gpt).
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
|