UndiMix-v3-13B: A Versatile Merged Language Model

UndiMix-v3-13B is a 13 billion parameter language model developed by Undi95, representing a personal mix designed for highly adaptable conversational interactions. This iteration, an evolution from its V2 predecessor, directly uses ReMM-S-Kimiko-v2-13B as its foundational base, moving away from Llama-2-13B-fp16.

Key Characteristics & Merging Strategy

The model's unique capabilities stem from a strategic SLERP merge of several distinct models, each contributing to its diverse output:

• Base Model: Undi95/ReMM-S-Kimiko-v2-13B (0.272 weight)
• Contributing Models:
  • The-Face-Of-Goonery/Huginn-13b-v1.2 (0.264 weight)
  • Doctor-Shotgun/llama-2-13b-chat-limarp-v2-merged (0.264 weight)
  • jondurbin/airoboros-l2-13b-2.1 (0.10 weight)
  • IkariDev/Athena-v1 (0.10 weight)

This specific blend allows UndiMix-v3 to generate responses that can be "hot, serious, playful," and effectively incorporate emojis, a feature enhanced by the inclusion of llama-2-13b-chat-limarp-v2-merged.

Prompt Template

The model utilizes the Alpaca prompt template, structured as follows:

Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{prompt}

### Response:

Ideal Use Cases

UndiMix-v3-13B is particularly well-suited for applications requiring:

• Dynamic conversational agents: Where the tone and style of interaction need to vary significantly.
• Creative content generation: For scenarios demanding expressive and emotionally nuanced text.
• Role-playing and interactive storytelling: Its ability to adapt to different personas and use emojis makes it suitable for engaging narrative experiences.