Nam-toon-studio/amrit-qwythos-corrections

Amrit-Qwythos Corrections Dataset 49 question → correct-answer pairs used to fine-tune Qwythos-9B (a Qwen3.5-based model) to fix specific, verified hallucinations caught during real use in the Amrit OS project. What this fixes Two real, reproduced failure classes: Self-identity confusion — at higher sampling temperature, the base model would sometimes answer "I am Qwen3.5, made by Alibaba Cloud" instead of its actual fine-tuned identity (Qwythos, by Empero AI).… See the full description on the dataset page: https://huggingface.co/datasets/Nam-toon-studio/amrit-qwythos-corrections.