Model Overview

The sthenno-com/miscii-14b-1028 is a 14.8 billion parameter language model with an extensive context length of 131072 tokens. Its core differentiator lies in its role-based instruction following mechanism, which allows developers to define distinct personas for both the user and the assistant directly within the system prompt. This enables highly customized and context-aware interactions.

Key Capabilities & Features

• Role-Based Instruction Following: Utilizes a unique system prompt structure (<|context_start|>personas<|context_sep|>...) to establish detailed user and assistant personas, facilitating nuanced conversational control.
• Extended Context Window: Supports a 131072-token context, enabling the processing and generation of very long sequences of text while maintaining coherence.

Performance Highlights

Evaluations on the Open LLM Leaderboard (Refined) indicate strong performance in key areas:

• IFEval (0-Shot): 82.37
• BBH (3-Shot): 49.26
• MATH Lvl 5 (4-Shot): 50.30
• MMLU-PRO (5-Shot): 46.14

These scores suggest the model is proficient in understanding and executing instructions, handling complex reasoning, and demonstrating mathematical capabilities.

Ideal Use Cases

This model is particularly well-suited for applications requiring:

• Advanced Role-Playing Scenarios: Where precise persona adherence and conversational style are critical.
• Complex Instruction Following: For tasks that demand strict interpretation and execution of detailed directives.
• Long-Context Applications: Benefiting from its large context window for summarizing, analyzing, or generating extensive documents.