Models

Open models for measurable RAG behavior.

The model collection documents task-specific fine-tunes across architectures and sizes, including intended use, limitations and evaluation context.

Task-specific fine-tunes across several Qwen2.5 model sizes.

Experiments with compact and medium Gemma models for grounded and structured responses.

Cross-family comparison models trained and evaluated with the same task design.

Model repository

Current releases, files, model cards and usage information are hosted on Hugging Face.