The sasa2000/Nemotron-Terminal-14B-heretic is a 14 billion parameter, 32K context length language model, fine-tuned from NVIDIA's Nemotron-Terminal-14B using the Heretic v1.2.0 tool to create a decensored version. Specialized for autonomous terminal interaction, this model is designed to generate structured JSON outputs for terminal commands and analysis. It demonstrates enhanced performance in terminal tasks compared to its Qwen3 base, with significantly reduced refusals compared to the original Nemotron-Terminal-14B.
Loading preview...
Model Overview
This model, sasa2000/Nemotron-Terminal-14B-heretic, is a 14 billion parameter variant of the Nemotron-Terminal family, fine-tuned from Qwen3 and specialized for autonomous terminal interaction. It has been processed with the Heretic v1.2.0 tool to produce a decensored version of the original nvidia/Nemotron-Terminal-14B model.
Key Capabilities & Features
- Autonomous Terminal Interaction: Specifically designed and fine-tuned for tasks involving terminal environments.
- Decensored Output: Modified to reduce refusals, achieving 64/100 refusals compared to the original model's 98/100.
- Structured JSON Output: Generates responses in a structured JSON format, including analysis, plans, and executable commands with keystrokes and durations.
- Enhanced Terminal Performance: Demonstrates significant improvements in terminal task accuracy, achieving 20.2% on Terminal-Bench 2.0 compared to the Qwen3 base accuracy of 4.04% for the 14B variant.
- Terminus 2 Scaffolding: Trained and optimized for use with Terminus 2 scaffolding for consistent evaluation and deployment.
Use Cases
This model is particularly well-suited for applications requiring:
- Automated execution of terminal commands.
- Intelligent agents for system administration or development workflows.
- Generating structured plans and analyses for complex terminal tasks.
- Scenarios where a less restrictive language model is preferred for terminal-based interactions.