l3utterfly/phi-2-layla-v1
TEXT GENERATIONConcurrency Cost:1Model Size:3BQuant:BF16Ctx Length:2kPublished:Mar 1, 2024License:mitArchitecture:Transformer0.0K Open Weights Warm
l3utterfly/phi-2-layla-v1 is a 3 billion parameter Phi-2 model, fine-tuned by l3utterfly using the OpenHermes 2.5 dataset. This model is optimized for multi-turn conversational tasks and character impersonation, making it suitable for interactive applications. It features a 2048-token context length and is licensed under MIT.
Loading preview...
Model Overview
l3utterfly/phi-2-layla-v1 is a 3 billion parameter language model based on Microsoft's Phi-2 architecture, developed by l3utterfly and funded by Layla Network. It has been specifically fine-tuned using a pre-processed version of the OpenHermes 2.5 dataset.
Key Capabilities
- Multi-turn Conversation: Optimized for engaging in extended, multi-turn dialogues.
- Character Impersonation: Designed to effectively adopt and maintain specific character personas.
- Refusal Handling: The training data was processed to remove refusals, aiming for more direct responses.
- NSFW Content Generation: Includes NSFW generated conversations from the Teatime dataset, indicating a broader content generation capability.
Training Details
The OpenHermes 2.5 dataset underwent specific pre-processing steps:
- Removal of all refusal-based responses.
- Elimination of mentions of "AI assistant."
- Splitting multi-turn dialogues into individual conversational records.
- Integration of NSFW conversations from the Teatime dataset.
Good For
- Offline Personal Assistants: Serves as the base model for Layla, an offline personal assistant.
- Interactive Applications: Ideal for scenarios requiring dynamic, multi-turn interactions.
- Role-playing and Character-driven AI: Excels in applications where the AI needs to maintain a specific character or persona.