Hugging Face Overview
Amadeus-Verbo-FI-Qwen2.5-3B-PT-BR-Instruct Overview
Amadeus-Verbo-FI-Qwen2.5-3B-PT-BR-Instruct is a specialized large language model (LLM) developed by amadeusai, focusing on the Brazilian Portuguese language. It is built upon the robust Qwen2.5-3B-Instruct base model, undergoing a targeted fine-tuning process over two epochs using a substantial 600,000-instruction dataset.
Key Capabilities & Technical Details
- Language Specialization: Primarily designed and optimized for Brazilian Portuguese (PT-BR).
- Architecture: Utilizes a Transformer-based architecture incorporating features like RoPE, SwiGLU, RMSNorm, and Attention QKV bias.
- Parameter Count: Features 3.09 billion parameters, with 2.77 billion non-embedding parameters.
- Context Length: Supports a significant context window of 32,768 tokens, enabling processing of longer inputs and generating more coherent responses.
- Fine-tuning: Enhanced for instruction-following tasks through fine-tuning on a large instruction dataset.
Good For
- Brazilian Portuguese Applications: Ideal for applications requiring high-quality text generation and understanding in Brazilian Portuguese.
- Instruction Following: Excels at responding to specific instructions and prompts due to its instruction-tuned nature.
- Research and Development: Suitable for researchers and developers working on PT-BR NLP tasks, offering a strong base model with a substantial context window. Further technical details are available in the associated arXiv paper.