Dilemma-5

Dilemma-5 is an independent, open-source web application engineered to run optimized LLMs locally on consumer hardware. Developed as both a technical exploration and a italian joke about a new SLM, the project serves as a dedicated web interface for the Italian-centric EMMA-5 model (and other ONNX-converted architectures). Built with a lightweight Python and Flask backend, it delivers a clean, dark-themed interface that allows users to perform low-latency text generation in complete privacy, detached from external APIs or data dependencies.

The entire deployment pipeline is fully automated via scripting, streamlining dependency installation and virtual environment isolation to ensure a one-click setup.

Key Features:

• EMMA-5 & ONNX Optimization: Specifically tailored to run the EMMA-5 model converted to the ONNX format.
• Hardware Acceleration via CUDA: Native support for NVIDIA CUDA 12 via ONNX Runtime's Execution Providers.
• Smart CPU Fallback: Built-in safety that automatically falls back to CPU execution if compatible NVIDIA drivers or hardware are missing.
• Automated Isolated Deployment: Includes custom automated batch script (.bat) to handle local Python virtual environments (venv) and dependency installation.
• 100% Air-Gapped & Private: Designed to operate entirely offline, ensuring data privacy by keeping prompts and weights strictly on local storage.