|
--- |
|
license: mit |
|
datasets: |
|
- merve/turkish_instructions |
|
language: |
|
- tr |
|
--- |
|
|
|
![image/png](https://cdn-uploads.huggingface.co/production/uploads/6324eabf05bd8a54c6eb1650/2tIft9E7Ef2XU97X_zjfE.png) |
|
|
|
This QLoRA was trained for 0.25 epochs on TinyLlama 1.1B with Merve's Turkish_instructions dataset. It aims to take advantage of TinyLlama's huge token count. |
|
|
|
It is exceptionally bad (due to low amount of training), but it can consistently produce meaningful words and rarely form coherent sentences in Turkish. |
|
|
|
You can use the default template for TinyLlama. |
|
|
|
--------------------------------------------- |
|
|
|
Bu QLoRA Merve'nin Turkish_instructions veri setiyle 0.25 epoch eğitildi. TinyLlama'nın yüksek token sayısından yararlanmayı amaçlamaktadır. |
|
|
|
Sonuçlar kötü (daha fazla eğitim gerek) ama kelime boyutunda anlamlı sonuçlar verip bazen anlamlı Türkçe cümleler kurabiliyor. |
|
|
|
TinyLlama chat şablonuyla kullanabilirsiniz. |