|
|
--- |
|
|
base_model: Qwen/Qwen3-4B |
|
|
datasets: |
|
|
- LucidityAI/Astral-Post-Training-Dataset |
|
|
tags: |
|
|
- code |
|
|
- chemistry |
|
|
- finance |
|
|
- biology |
|
|
--- |
|
|
|
|
|
# Astral-4B |
|
|
|
|
|
Astral 4B is the medium sized model in the Astral family. It was fine-tuned from Qwen3 4b on LucidityAI/Astral-Post-Training-Dataset. |
|
|
|
|
|
|
|
|
As with usual Qwen3 models, reasoning can be toggled through the usage of ```/no_think``` or not. |
|
|
|
|
|
|
|
|
### Example Prompt (ChatML Format (THINKING)): |
|
|
|
|
|
```xml |
|
|
<|im_start|>user |
|
|
What is the capital of France? |
|
|
<|im_end|> |
|
|
<|im_start|>assistant |
|
|
<think> |
|
|
``` |
|
|
|
|
|
### Example Prompt (ChatML Format (NON-THINKING)): |
|
|
|
|
|
```xml |
|
|
<|im_start|>user |
|
|
What is the capital of France? /no_think |
|
|
<|im_end|> |
|
|
<|im_start|>assistant |
|
|
<think> |
|
|
``` |
