Deepseek or something else?
This is likely a runtime configuration adjustment made by Moderno Tech when packaging/quantizing, rather than retraining the model. DeepSeek-Coder-V2-Lite itself supports context extension via YaRN , and 160K is likely an extension based on this?!?
Yes, that's correct. Nova-Coder is a model developed and owned by Moderno Tech, with DeepSeek-Coder-V2-Lite serving as its foundation.
I personally lead the development, customization, distillation, and fine-tuning process. I've adapted the model specifically for my development workflows, Moderno Tech projects, and agentic programming tasks.
This is not simply a renamed or repackaged DeepSeek model. My work includes targeted distillation, specialized fine-tuning, runtime optimization, and customization of its behavior to better support autonomous coding agents and advanced software development workflows.
Another important aspect of my approach is uncensored AI development. I've fine-tuned the model to reduce unnecessary content restrictions and provide greater freedom and flexibility in legitimate technical, creative, and programming tasks. My philosophy is to prioritize user control and minimize artificial limitations without compromising technical capabilities.
The current release is an intermediate stage in a much larger project: building an independent AI ecosystem under Moderno Tech.
My ultimate goal for the next major generation of Nova-Coder is to develop a completely original model from scratch, including its own architecture, training pipeline, and model weights, without relying on an existing foundation model.
DeepSeek is the starting point of this particular release. The customization, development work, technical direction, and long-term vision are driven by Moderno Tech.