Polygl0t/Tucano2-qwen-3.7B-Instruct
Polygl0t/Tucano2-qwen-3.7B-Instruct is a 3.76 billion parameter instruction-tuned Portuguese language model built on a Transformer-based Qwen3 architecture. Developed by Polygl0t, it was trained using supervised fine-tuning and Anchored Preference Optimization. This model excels in Portuguese benchmarks and supports tasks like RAG, function calling, summarization, and structured output generation, making it suitable for research and development in Portuguese language modeling.
Loading preview...
Model Overview
Polygl0t/Tucano2-qwen-3.7B-Instruct is a 3.76 billion parameter instruction-tuned Portuguese language model, developed by Polygl0t. It is built upon the Transformer-based Qwen3 architecture and was trained using a combination of supervised fine-tuning (SFT) and Anchored Preference Optimization (APO). The model has a context length of 4,096 tokens.
Key Capabilities
- Portuguese Language Proficiency: Optimized for a wide range of Portuguese benchmarks.
- Diverse Task Support: Capable of retrieval-augmented generation (RAG), function calling, tool use, summarization, and structured output generation.
- Open and Reproducible: All datasets, source code, and training recipes for the Tucano2 series are publicly available.
Performance Highlights
Evaluations show Tucano2-qwen-3.7B-Instruct achieving a Total Average NPM of 53.64, with strong performance in Knowledge & Reasoning (56.22 NPM) and competitive scores in Instruction Following (41.67) and Coding (47.56). It demonstrates robust performance against other models in its size class, particularly in Portuguese-specific tasks.
Intended Uses
This model is primarily intended for research and development in Portuguese language modeling. It can also serve as a foundation for further fine-tuning and adaptation, provided usage adheres to the Apache 2.0 license. Users are advised to conduct their own risk and bias assessments for real-world applications.