Model Card for hacer201145/Hasex0.2-0.6B

Model Details

Description

Here comes Hasex0.2-0.6B. I deleted the previous version because it was worse. I also decided to change the base model from DialoGPT-medium (this is not an assistant, but rather a meme generator). But, okay. It was trained on ~9500 qualitative examples (including algebra, geometry, etc.). In the results, it shows better speed and responses, unlike its base model Qwen3-0.6B. Although it was trained in only one era, reasoning has become shorter in it, and it is more of an immediate essence than endless "hmm". I started collecting the dataset on Friday and finished it yesterday. After that, I was still tormented by how to get it, but it was definitely worth the result. To use it, I recommend using the example from the Qwen3-0.6B model card. It is very well represented there. But okay, I decided to take a break from it a bit, so expect Hasex0.3 in at least 3-4 months. It will already use this model as the main one. And I want to make a dataset of ~20,000 examples for it. Okay, thanks to everyone who read this.

• Developed by: hacer201145
• License: Apache license 2.0
• Finetuned from model Qwen3-0.6B

Model Sources

• Repository: Maybe soon?
• Paper [optional]: Soon
• Demo [optional]: Maybe soon?