Llama_3.1_8b_DodoWild_v2.01
| Property | Value |
|---|---|
| Base Model | LLaMA 3.1 8B |
| Merge Method | Model Stock |
| Model URL | Hugging Face |
| Average Benchmark Score | 30.31 |
What is Llama_3.1_8b_DodoWild_v2.01?
Llama_3.1_8b_DodoWild_v2.01 is a sophisticated merge of multiple LLaMA 3.1-based models, created by Nexesenex using mergekit. Built upon the SentientAGI/Dobby-Mini-Unhinged-Llama-3.1-8B base, it combines the strengths of Dolermed and Smarteaz variants to create a more capable language model.
Implementation Details
The model utilizes the Model Stock merge method with equal weights (1.0) assigned to both Dolermed and Smarteaz variants. It employs bfloat16 dtype for efficient computation and includes normalized weights with an auto chat template.
- Custom tokenizer configuration using union method
- Normalized weight distribution across merged models
- Built on LLaMA 3.1 8B architecture
Core Capabilities
- Strong instruction-following capability (IFEval: 79.78)
- Decent performance on BBH 3-shot tasks (32.11)
- Mathematical reasoning capabilities (MATH Lvl 5: 19.86)
- Professional knowledge evaluation (MMLU-PRO: 30.43)
Frequently Asked Questions
Q: What makes this model unique?
This model's uniqueness lies in its balanced merge of two specialized LLaMA 3.1 variants, optimized for instruction following while maintaining general-purpose capabilities.
Q: What are the recommended use cases?
Given its strong performance in instruction following and reasonable scores across various benchmarks, this model is well-suited for general-purpose tasks, particularly those requiring precise instruction following and professional domain knowledge.
