Undi95/ReMM-v2.2-L2-13B: A MythoMax Recreation

Undi95/ReMM-v2.2-L2-13B is a 13 billion parameter model that serves as an updated recreation of the original MythoMax-L2-B13. It leverages the SLERP merging method to combine several prominent base models, aiming to integrate their strengths into a single, cohesive unit.

Key Components and Merging Strategy

This model is a merge of:

• ReML-v2.2: A composite of The-Face-Of-Goonery/Chronos-Beluga-v2-13bfp16, jondurbin/airoboros-l2-13b-2.2.1, and NousResearch/Nous-Hermes-Llama2-13b.
• The-Face-Of-Goonery/Huginn-13b-v1.2.

The merging process specifically updates the Airoboros component to its latest version (2.2.1) and incorporates the Huginn model, which is noted as a "hottest" component, suggesting recent development or strong performance.

Performance Snapshot

Evaluated on the Open LLM Leaderboard, ReMM-v2.2-L2-13B achieves an average score of 50.45. Notable individual benchmark results include:

• ARC (25-shot): 61.26
• HellaSwag (10-shot): 84.16
• MMLU (5-shot): 56.22
• TruthfulQA (0-shot): 51.35
• Winogrande (5-shot): 75.61

Prompt Format

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:

Good for:

• General-purpose language generation and understanding tasks.
• Users seeking a merged model that combines the characteristics of Chronos-Beluga, Airoboros, Nous-Hermes, and Huginn.
• Experimentation with models built using the SLERP merging technique.