Overview

CabraLlama3-70b is a 70 billion parameter language model developed by botbot-ai, based on the Meta Llama 3 70B Instruct architecture. This model has undergone further refinement using the proprietary Cabra 30k dataset, with a primary focus on optimizing its performance for the Portuguese (pt-br) language.

Key Capabilities & Features

• Portuguese Language Optimization: Specifically fine-tuned to understand and generate responses in Brazilian Portuguese.
• Llama 3 Base: Leverages the robust Llama 3 architecture, which is known for strong performance in dialogue use cases.
• Instruction-Tuned: Optimized for instruction-following and conversational applications through supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF).
• Quantized Versions: Offers various quantized GGUF versions for efficient deployment, available in the "quantanization" branch.

Performance

The model demonstrates strong performance on the Open Portuguese LLM Leaderboard, achieving an average score of 78.44. Notable results include:

• ENEM Challenge (No Images): 82.02
• Assin2 RTE: 93.21
• Assin2 STS: 83.32
• FaQuAD NLI: 80.60

Use Cases

This model is particularly well-suited for applications requiring high-quality, nuanced interactions in Portuguese, such as:

• Chatbots and conversational AI for Portuguese-speaking users.
• Content generation in Portuguese.
• Language understanding and analysis tasks specific to Portuguese.