bofenghuang/vigogne-7b-instruct
Vigogne-7B-Instruct by bofenghuang is a 7 billion parameter LLaMA-based instruction-following model specifically fine-tuned for the French language. It is designed to understand and respond to instructions in French, leveraging an expanded training dataset of 262k examples. This model is primarily intended for research purposes in French natural language processing applications.
Loading preview...
Vigogne-7B-Instruct: A French Instruction-following LLaMA Model
Vigogne-7B-Instruct is a 7 billion parameter LLaMA model developed by bofenghuang, specifically fine-tuned to follow instructions in French. This model is built upon the LLaMA architecture and has undergone several iterations of improvement, primarily focusing on expanding and refining its French instruction-following dataset.
Key Capabilities
- French Instruction Following: Optimized to understand and generate responses based on instructions provided in French.
- LLaMA-7B Base: Leverages the robust LLaMA-7B architecture for its foundational language understanding.
- Iterative Dataset Expansion: Training dataset has been progressively expanded from an initial translated Stanford Alpaca dataset to 262k examples, enhancing performance and instruction adherence.
- Research-Oriented: Intended and licensed for research use only, aligning with the Stanford Alpaca project's licensing.
When to Use This Model
Vigogne-7B-Instruct is particularly well-suited for:
- French NLP Research: Ideal for academic and research projects requiring a capable instruction-following model in French.
- Exploring Multilingual Fine-tuning: Useful for researchers studying the effects of fine-tuning large language models for specific non-English languages.
- Developing French-centric Applications: Can serve as a base for experimental applications that require understanding and generating French text based on user instructions.
Limitations
As an ongoing development, Vigogne-7B-Instruct may still produce harmful, biased, or incorrect content. Its use is restricted to research purposes due to its CC BY NC 4.0 license.