lucianosb/boto-9B-it

Overview

Boto 9B IT is a 9 billion parameter instruction-tuned language model developed by lucianosb, specifically fine-tuned for the Portuguese language. It is based on the Gemma2-9B-it architecture and was trained using the cetacean-ptbr dataset. The model was developed efficiently using Unsloth and Huggingface's TRL library.

Key Characteristics

• Language Focus: Primarily designed and optimized for Portuguese.
• Response Style: Tends to generate verbose and detailed responses.
• Base Model: Fine-tuned from unsloth/gemma-2-9b-it-bnb-4bit.
• Performance: Achieves an average score of 73.65 on the Open Portuguese LLM Leaderboard, with notable scores such as 94.04 on Assin2 RTE and 82.85 on HateBR Binary.

Important Considerations

• Content Moderation: The model does not include built-in content moderation mechanisms and may reproduce stereotypes or generate content inconsistent with reality due to biases in its training data.
• Reliability: Users are advised not to rely exclusively on the model for critical decisions and to exercise their own judgment when interpreting generated content.

Use Cases

This model is well-suited for applications requiring extensive and detailed text generation in Portuguese, particularly where a verbose output style is acceptable or desired. Its strong performance on Portuguese-specific benchmarks suggests utility in various language-centric tasks.