Overview

This model, grimjim/Llama-3-Instruct-8B-SimPO-SPPO-Iter3-merge, is an 8 billion parameter instruction-tuned language model built upon the Meta Llama 3 architecture. It was created by merging two base models: princeton-nlp/Llama-3-Instruct-8B-SimPO and UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3, using the SLERP merge method via mergekit.

Key Capabilities & Performance

• Merged Architecture: Combines the strengths of two distinct Llama 3 instruction-tuned models.
• Text Generation: Primarily designed for text generation tasks.
• Evaluated Benchmarks: Performance metrics are available on the Open LLM Leaderboard, including:
  • IFEval (0-Shot): 68.06 strict accuracy
  • BBH (3-Shot): 29.07 normalized accuracy
  • MATH Lvl 5 (4-Shot): 6.19 exact match
  • MMLU-PRO (5-shot): 29.83 accuracy

When to Use This Model

• General Instruction Following: Suitable for tasks requiring a merged Llama 3 instruction-tuned base.
• Exploration of Merged Models: Ideal for developers interested in the performance characteristics of models created via SLERP merging of specific Llama 3 variants.
• Benchmarking: Can be used for further evaluation against the provided Open LLM Leaderboard results.