kingbri/Nous-Hermes-limarp-l2-13B
kingbri/Nous-Hermes-limarp-l2-13B is a 13 billion parameter language model, a merge of Nous Hermes Llama 2 13b and LimaRP llama 2 Lora. This model combines the strong reasoning and instruction-following capabilities of Nous Hermes with the character roleplay strengths of LimaRP. It is specifically designed for nuanced character interaction and roleplaying scenarios, offering a unique blend of general instruction adherence and specialized conversational depth.
Loading preview...
Model Overview
kingbri/Nous-Hermes-limarp-l2-13B is a 13 billion parameter model created by kingbri through a strategic merge of two distinct models: Nous Hermes Llama 2 13b and LimaRP llama 2 Lora. The merge was performed using ez-trainer by CoffeeVampire/Blackroot, with the LimaRP LoRA integrated at a weight of 0.66 to preserve character personality nuances.
Key Capabilities
- Blended Strengths: This model aims to leverage the robust reasoning and instruction-following abilities of Nous Hermes Llama 2 with the enhanced character roleplay capabilities of LimaRP. This makes it suitable for applications requiring both structured responses and dynamic, personality-driven interactions.
- Roleplay Optimization: The specific merge ratio for LimaRP was chosen to prevent the LoRA from overpowering the base model, ensuring that character personalities remain distinct and nuanced.
- Instruction Format Flexibility: It supports both Alpaca 2 and LimaRP instruction formats, allowing users to choose the most appropriate prompt structure for their specific use case.
Intended Use Cases
This model is particularly well-suited for:
- Character Roleplaying: Excelling in scenarios where detailed character interaction and personality consistency are crucial.
- Interactive Storytelling: Generating responses that maintain character voice and narrative flow.
- Combined Instruction and Roleplay: Use cases that require both adherence to instructions and engaging, character-driven dialogue.
Limitations
Users should be aware that the model may exhibit biases similar to those found in niche online roleplaying communities. It is not intended for providing factual information or advice due to its specialized training for creative and interactive content.