datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
emotion-combined-trajectories-constant-control-gemma-4-31b-it
Per-token emotion trajectories, constant-emotion control
The control condition: google/gemma-4-31b-it read over stories where the scene changes but the emotion does not. A measure that reports a transition here is responding to the scene change, not to emotion.
Each story is stored as one .npz. The arrays are per token, so a trajectory
can be replayed word by word rather than only summarised.
Contents
Path
Contents
shards/<story_id>.npz
one story… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-combined-trajectories-constant-control-gemma-4-31b-it.emotion-probe-controls
Emotion probe controls
Synthetic expression-story datasets for extracting and comparing emotion and
bodily-state vectors from large language models. Each story portrays a character
experiencing a target feeling or state in a specified scenario, without naming
the target label or its direct synonyms.
The repository separates shared generation inputs from task-specific
labels, prompts, and stories. Bodily-state stories are available now; the
emotion-expression folder currently… See the full description on the dataset page: https://huggingface.co/datasets/yaelwalker/emotion-probe-controls.emotion-combined-stories-deepseek-v4-pro-constant-control
Also available, browsable: this corpus is one source inside
abotresol/emotion-story-corpora,
which unions every story corpus from this project into two Parquet subsets
with a working dataset viewer. This repository stays as the original,
unchanged, so existing links and citations keep resolving.
license: mit
task_categories:
- feature-extraction
tags:
- interpretability
- emotion
- gemma
- ai-safety
Constant-emotion control stories
Control stories in… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-combined-stories-deepseek-v4-pro-constant-control.emotion-combined-trajectories-constant-control-gemma-4-31b-it-webdataset
emotion-combined-trajectories-constant-control-gemma-4-31b-it-webdataset
The same per-token trajectory data as
abotresol/emotion-combined-trajectories-constant-control-gemma-4-31b-it, packed as
WebDataset tars instead of
209 separate .npz files.
Nothing changed but the packaging. Each tar member is the original .npz,
byte for byte; the build script checks every sample's SHA-256 against the file
it came from before publishing. The original repository stays where it is.
Why: 209… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-combined-trajectories-constant-control-gemma-4-31b-it-webdataset.