Model Overview

The rombodawg/rombos_Replete-Coder-Instruct-8b-Merged is an 8 billion parameter instruction-tuned model developed by rombodawg. It is a TIES merge of two base models: meta-llama/Meta-Llama-3-8B-Instruct and Replete-AI/Llama3-8B-Instruct-Replete-Adapted. This merging strategy aims to combine the strengths of both foundational models.

Key Capabilities & Performance

This merged model reportedly shows improved performance in both coding tasks and overall general capabilities compared to its individual base models. While specific benchmarks are still forthcoming, initial evaluations on the Open LLM Leaderboard indicate:

• Average Score: 16.32
• IFEval (0-Shot): 53.88
• BBH (3-Shot): 21.94
• MATH Lvl 5 (4-Shot): 7.10
• GPQA (0-shot): 2.57
• MuSR (0-shot): 3.45
• MMLU-PRO (5-shot): 8.98

Use Cases

Given its reported enhancements in coding and general performance, this model is particularly suited for applications requiring robust code generation, understanding, and other instruction-following tasks where a balance of general intelligence and specialized coding ability is beneficial. Its 8192 token context length supports processing moderately long inputs.