mshojaei77/gemma-3-4b-persian-v0

Model Overview

mshojaei77/gemma-3-4b-persian-v0 is a 4.3 billion parameter language model built on the Gemma 3 architecture, specifically fine-tuned for the Persian language. It utilizes QLoRA for 4-bit quantization, significantly reducing computational overhead while maintaining strong performance in Persian text generation and comprehension. A notable feature is its inherited capability for image input, alongside its primary focus on Persian language tasks.

Key Capabilities

• Persian Language Specialization: Optimized for generating and understanding Persian text.
• Efficient Deployment: Uses QLoRA 4-bit quantization for reduced memory footprint and faster inference.
• Multimodal Input: Retains image input capabilities from the base Gemma 3 model.
• Instruction Following: Designed to interpret and execute text-based instructions in Persian.

Training Details

The model was fine-tuned using Supervised Fine-Tuning (SFT) on the mshojaei77/Persian_sft dataset, which comprises approximately 681,000 rows of Persian text focused on instruction-following and conversational interactions. This process was conducted on a single T4 GPU, utilizing Hugging Face Transformers, peft, and bitsandbytes.

Intended Use Cases

• Question Answering: Accurately responding to Persian language queries.
• Text Generation: Producing fluent and context-aware Persian content.
• Conversational AI: Integration into chatbots and virtual assistants for Persian speakers.

Limitations

Users should be aware of potential limitations including reduced precision due to 4-bit quantization, the absence of comprehensive evaluation metrics specific to this variant, and the possibility of biases inherited from training data. The model has not undergone safety tuning, requiring caution in sensitive applications.