Vigogne-13B-Instruct: A French Instruction-Following LLaMA Model
Vigogne-13B-Instruct is a 13 billion parameter LLaMA model developed by bofenghuang, specifically fine-tuned to understand and follow instructions in French. This model is built upon the LLaMA architecture and has undergone several iterations of dataset expansion to enhance its performance in French instruction-following tasks.
Key Capabilities
- French Instruction Following: Optimized for processing and responding to prompts given in French.
- LLaMA-Based Architecture: Leverages the foundational capabilities of the LLaMA model family.
- Iterative Dataset Improvement: Trained on an progressively expanded dataset, reaching 262k translated Stanford Alpaca instructions, to improve translation quality and overall performance.
Good For
- Research in French NLP: Ideal for academic and research applications focusing on French language understanding and generation.
- Developing French-centric AI Applications: Suitable for experimental projects requiring a model proficient in French instruction adherence.
- Exploring LLaMA's Multilingual Adaptations: Useful for researchers studying the adaptation of large language models to specific non-English languages.
Limitations
Vigogne-13B-Instruct is still under development and may generate harmful, biased, incorrect, or unhelpful content. It is licensed for research use only, consistent with the Stanford Alpaca project.