metadata
base_model: LucidityAI/Astral-4B
datasets:
- LucidityAI/Astral-Post-Training-Dataset
tags:
- code
Astral-4B-Coder
Astral 4B is the largest sized model in the Astral coder family. It was fine-tuned from Astral 4b on LucidityAI/Astral-Post-Training-Dataset.
Use /no_think for agentic tasks, use thinking for non-agentic coding tasks/
As with usual Qwen3 models, reasoning can be toggled through the usage of /no_think or not.
Example Prompt (ChatML Format (THINKING)):
<|im_start|>user
What is the capital of France?
<|im_end|>
<|im_start|>assistant
<think>
Example Prompt (ChatML Format (NON-THINKING)):
<|im_start|>user
What is the capital of France? /no_think
<|im_end|>
<|im_start|>assistant
<think>