AyoubChLin/Qwen3.5-9B-saudi-dialect
AyoubChLin/Qwen3.5-9B-saudi-dialect is a 9 billion parameter language model fine-tuned from unsloth/Qwen3.5-9B, specifically optimized for generating chat in Saudi Arabian dialect. It was trained using Unsloth LoRA SFT on a dataset of Saudi Arabic conversations, making it distinct for its specialized regional linguistic capability. This model is designed for direct use with plain transformers, offering a maximum sequence length of 4096 tokens for dialect-specific conversational AI.
Loading preview...
Model Overview
AyoubChLin/Qwen3.5-9B-saudi-dialect is a specialized 9 billion parameter language model, fine-tuned from the unsloth/Qwen3.5-9B base model. Its primary distinction is its fine-tuning for generating text in the Saudi Arabian dialect, making it highly suitable for applications requiring localized Arabic communication. The model was trained using Unsloth LoRA SFT on a dedicated dataset of Saudi Arabic conversations, and the full weights are merged into a standalone checkpoint for direct deployment.
Key Capabilities
- Saudi Dialect Generation: Excels at producing conversational text in Saudi Arabic, with a system prompt configured for this specific dialect.
- Qwen3.5 Architecture: Leverages the robust Qwen3.5 base for its underlying language understanding and generation capabilities.
- Optimized for Chat: Designed for chat-based interactions, as indicated by its training on conversational data.
- Standard Integration: Usable directly with the
transformerslibrary, providing straightforward integration for developers.
Training Details
The model was fine-tuned on the HeshamHaroon/saudi-dialect-conversations dataset, comprising 3366 training examples. Training involved 4 epochs with a learning rate of 4e-4, utilizing an NVIDIA A100-SXM4-80GB GPU. The final eval/loss was 1.41955, indicating effective learning on the dialectal nuances.
Limitations
- No External Benchmarks: Performance is evaluated based on training/evaluation loss, without external benchmark comparisons.
- Dialectal Nuance: While specialized, coverage of Saudi dialect across all regions, phrasing styles, and topics may be uneven.
- Potential for Hallucination: Like other LLMs, it may still hallucinate or occasionally drift towards more formal Arabic depending on the input prompt.