---
language:
- ar
- en
pipeline_tag: text-generation
tags:
- arabic
- hayula-rushd
- qwen2
- instruct
- hayula
- mlx
license: apache-2.0
---

# Hayula-Rushd-Q-7B

> 🌟 **The rebranded flagship model of Hayula Labs** — formerly Averroes-Q-Instruct.

Arabic instruction-tuned model on 55K human-curated data, rebranded to establish a clean identity under **Hayula Labs** without naming conflicts.

**Base:** Qwen2.5-7B-Instruct
**Data:** 55,174 human-curated Arabic instruction-conversation pairs
**LoRA:** rank=8, scale=0.5, lr=1e-5, iters=5000
**Train Loss:** 1.434 (final)
**Training:** ~45 min on M2 Ultra

## Why the rebrand?

The Averroes name had naming conflicts with unrelated models on Hugging Face. Hayula-Rushd-Q-7B (رُشد = guidance/maturity) preserves the intellectual roots of Averroes (Ibn Rushd) while establishing a clean, independent identity under Hayula Labs.

## Usage



## Links
- 🔄 **Old repo (redirect):** [hayulalab/Averroes-Q-Instruct](https://huggingface.co/hayulalab/Averroes-Q-Instruct)
- 🏠 **Hayula Labs:** [hayula.xyz](https://hayula.xyz)