alifzl/zhaav-gemma3-4B
The alifzl/zhaav-gemma3-4B model is a 4.3 billion parameter, Persian-specific language model fine-tuned on the Gemma 3 architecture. It utilizes QLoRA 4-bit quantization to optimize for generating and understanding Persian text efficiently. This model is designed for strong performance on commodity hardware without dedicated GPUs, making it suitable for resource-constrained environments. Its primary strength lies in instruction-following and conversational interactions within the Persian language.
Loading preview...
Model Overview
The alifzl/zhaav-gemma3-4B is a 4.3 billion parameter language model specifically fine-tuned for the Persian language. Built upon the Gemma 3 architecture, this model leverages QLoRA with 4-bit quantization to significantly reduce computational demands. This optimization allows it to deliver robust performance in generating and understanding Persian text, even on commodity hardware without dedicated GPUs.
Key Capabilities & Features
- Persian Language Specialization: Fine-tuned extensively on a Persian instruction-following and conversational dataset.
- Resource-Efficient: Utilizes QLoRA 4-bit quantization, making it suitable for deployment on systems with limited computational resources.
- Gemma 3 Architecture: Benefits from the foundational capabilities of the Gemma 3 model family.
- Instruction-Following: Optimized for tasks requiring adherence to instructions and engaging in conversational interactions in Persian.
Training Details
The model was fine-tuned using Supervised Fine-Tuning (SFT) with QLoRA on the mshojaei77/Persian_sft dataset. This dataset comprises approximately 680,000 rows of Persian text focused on instruction-following and conversational exchanges.
Usage
This model is compatible with both the Hugging Face Transformers library and Ollama, providing flexible deployment options for developers working with Persian language tasks.