Model Overview
l3utterfly/tinyllama-1.1b-layla-v4 is a 1.1 billion parameter language model built upon the Llama2 architecture. Developed by l3utterfly and funded by Layla Network, this model is a fine-tuned version of TinyLlama 1.1B, specifically optimized for interactive conversational tasks.
Key Capabilities
- Multi-turn Conversation: The model is fine-tuned on the OpenHermes 2.5 dataset, with specific preprocessing steps to enhance its ability to handle and generate coherent multi-turn dialogues.
- Character Impersonation: It excels at embodying specific characters and personalities, making it suitable for role-playing and personalized assistant applications.
- Offline Personal Assistant: This model serves as the base for Layla, an offline personal assistant, indicating its design for local, responsive interactions.
Training Details
The OpenHermes 2.5 dataset used for fine-tuning underwent specific preprocessing:
- Refusals and mentions of "AI assistant" were removed.
- Multi-turn dialogues were split into individual conversation records.
- NSFW generated conversations from the Teatime dataset were added.
Use Cases
This model is particularly well-suited for applications requiring:
- Engaging and extended conversational interactions.
- Role-playing scenarios where the model needs to maintain a consistent persona.
- Integration into offline personal assistant systems, such as the Layla platform.