Model Overview
The rhaymison/Llama-3-portuguese-Tom-cat-8b-instruct is an 8 billion parameter instruction-tuned model based on the Llama 3 architecture, developed by rhaymison. Its primary goal is to fill the gap in high-quality Portuguese language models, having been extensively fine-tuned on a superset of 300,000 Portuguese chat conversations. This specialization makes it particularly adept at handling conversational tasks and understanding nuances of the Portuguese language.
Key Capabilities
- Portuguese Language Proficiency: Optimized for chat and general language understanding in Portuguese.
- Instruction Following: Designed to follow instructions effectively, especially in conversational contexts.
- Quantization Support: Supports 4-bit and 8-bit quantization for efficient deployment on less powerful hardware (e.g., T4 or V100 GPUs), with full model requiring A100.
- GGUF Compatibility: A GGUF family of this model (e.g., Llama-3-portuguese-Tom-cat-8b-instruct-q8-gguf) is available for use with LlamaCpp.
Performance Highlights
Evaluated on the Open Portuguese LLM Leaderboard, the model achieved an Average score of 70.57. Notable scores include:
- Assin2 RTE: 90.91
- HateBR Binary: 86.99
- FaQuAD NLI: 76.05
- OAB Exams: 51.07
Ideal Use Cases
- Portuguese Chatbots: Excellent for building conversational AI agents that interact in Portuguese.
- Portuguese Content Generation: Suitable for generating text, answering questions, and summarizing information in Portuguese.
- Educational Applications: Can act as a tutor or provide explanations in Portuguese, as demonstrated by its ability to explain mathematical concepts.
- Resource-Constrained Environments: Quantized versions allow for deployment on hardware with limited memory.