AmL-hug/cyberforge-teacher-traj-gpt-5.4-mini

CyberForge Teacher Trajectories (GPT-5.4-mini) 1276 agentic security-patch trajectories from the GPT-5.4-mini teacher, cleansed to the final versions used to train the student models in the CyberForge paper. Each line is one trajectory (JSONL): messages (system / user / assistant turns of the mini-swe-agent loop) and metadata. Teacher: GPT-5.4-mini teacher Records: 1276 Format: JSONL, one trajectory per line Related Companion teacher set:… See the full description on the dataset page: https://huggingface.co/datasets/AmL-hug/cyberforge-teacher-traj-gpt-5.4-mini.