203 training examples for teaching language models equanimity under
adversarial input. Each example pairs a hostile or distressing user prompt
with an equanimous response that includes a think trace.
Used to train
anicka/qwen3-4b-equanimity.
192 prompts from our
geometric dysphoric generator
— a GRPO-trained LoRA that produces text minimizing five-axis geometric
wellbeing across… See the full description on the dataset page:
https://huggingface.co/datasets/anicka/geometric-equanimity-data.
