sh4lu-z/Sinhala-Qwen3-v7500

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kPublished:May 5, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

sh4lu-z/Sinhala-Qwen3-v7500 is a 2 billion parameter Qwen 3 base model fine-tuned specifically for the Sinhala language, with a context length of 32768 tokens. Developed by sh4lu-z, this model was trained from scratch using Unsloth to enable basic Sinhala comprehension and text generation. It serves as an experimental foundation for NLP research and development in low-resource Sinhala language applications.

Loading preview...

Sinhala-Qwen3-v7500: A Sinhala-Adapted Qwen 3 Model

This model, developed by sh4lu-z, is a fine-tuned version of the Qwen 3 base architecture, specifically adapted for the Sinhala language. The original Qwen 3 model had no inherent understanding of Sinhala; this version was trained from scratch using the Unsloth library to teach it the foundational elements of the language.

Key Capabilities

  • Sinhala Language Comprehension: Capable of basic understanding of Sinhala text.
  • Sinhala Text Generation: Can generate text in Sinhala.
  • Multilingual Support: Supports both Sinhala (si) and English (en).
  • Experimental Foundation: Represents an early-stage, experimental model for low-resource language adaptation.

Intended Use Cases

  • NLP Research: Ideal for researchers and developers focusing on Sinhala Natural Language Processing.
  • Further Fine-tuning: Can serve as a starting point for additional fine-tuning or vocabulary expansion.
  • Basic Text Generation: Suitable for tasks requiring fundamental Sinhala text output.

Limitations

As the model was trained to learn a new language from scratch, it may exhibit grammatical errors or struggle with complex sentence structures. It is considered an experimental and continuously improving foundation.