Mansourizona/mira-darija-v1
Mansourizona/mira-darija-v1 is a 1.5 billion parameter Qwen2.5-based causal language model developed by Mansourizona. This model was fine-tuned using Unsloth and Huggingface's TRL library, enabling 2x faster training. It is designed for general language generation tasks, leveraging its efficient training methodology.
Loading preview...
Model Overview
Mansourizona/mira-darija-v1 is a 1.5 billion parameter language model developed by Mansourizona. It is built upon the Qwen2.5 architecture and has been fine-tuned to enhance its performance. The model leverages Unsloth for accelerated training, achieving a 2x speed improvement, and utilizes Huggingface's TRL library for its fine-tuning process.
Key Characteristics
- Base Model: Fine-tuned from
unsloth/Qwen2.5-1.5B-bnb-4bit. - Parameter Count: 1.5 billion parameters, offering a balance between performance and computational efficiency.
- Training Efficiency: Benefits from Unsloth's optimizations, resulting in significantly faster training times.
- License: Distributed under the Apache-2.0 license, allowing for broad usage and modification.
Potential Use Cases
This model is suitable for applications requiring a compact yet capable language model, especially where training speed and resource efficiency are important. Its Qwen2.5 foundation suggests general language understanding and generation capabilities.