datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
emotion-vectors-gemma-4-31b-it-postfix
Emotion vectors, google/gemma-4-31b-it (corrected extraction)
Residual-stream activations for google/gemma-4-31b-it, pooled per story and averaged per
emotion. Each emotion ends up as one direction in the model's activation space.
Read LINEAGE.md before using this. This set supersedes
abotresol/emotion-vectors-gemma-4-31b-it. The earlier extraction ran
while the tokenizer padded on the left, so the step that skips a story's first
50 tokens skipped padding instead. This set… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-vectors-gemma-4-31b-it-postfix.emotion-vectors-gemma-4-31b-it
Emotion vectors — gemma-4-31b-it (instruct) probed on the external gemma-4-4B story corpus
Data provenance (what made these activations)
Probed model: google/gemma-4-31b-it (instruct)
Input corpus: snae/emotion_stories_gemma_4_4B — stories written by gemma-4-4B, a smaller EXTERNAL model (generator is NOT the probed model)
Per-story pooled residual-stream activations and per-emotion mean vectors,
extracted with gemma4-emotion-vectors… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-vectors-gemma-4-31b-it.emotion-vectors-gemma-4-31b
Emotion vectors — gemma-4-31b (base) probed on the external gemma-4-4B story corpus
Data provenance (what made these activations)
Probed model (whose activations these are): google/gemma-4-31b (base)
Input corpus: snae/emotion_stories_gemma_4_4B — third-person emotion stories written by gemma-4-4B, a smaller EXTERNAL model (the open replication's published corpus; generator is NOT the probed model)
Per-story pooled residual-stream activations and per-emotion… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-vectors-gemma-4-31b.emotion-vectors-gemma-4-31b-postfix
Emotion vectors, google/gemma-4-31b (corrected extraction)
Residual-stream activations for google/gemma-4-31b, pooled per story and averaged per
emotion. Each emotion ends up as one direction in the model's activation space.
Read LINEAGE.md before using this. This set supersedes
abotresol/emotion-vectors-gemma-4-31b. The earlier extraction ran
while the tokenizer padded on the left, so the step that skips a story's first
50 tokens skipped padding instead. This set re-extracts… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-vectors-gemma-4-31b-postfix.emotion-selfstory-vectors-gemma-4-31b-it-postfix
Emotion vectors, google/gemma-4-31b-it (corrected extraction)
Residual-stream activations for google/gemma-4-31b-it, pooled per story and averaged per
emotion. Each emotion ends up as one direction in the model's activation space.
Read LINEAGE.md before using this. This set supersedes
abotresol/emotion-selfstory-vectors-gemma-4-31b-it. The earlier extraction ran
while the tokenizer padded on the left, so the step that skips a story's first
50 tokens skipped padding instead. This… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-selfstory-vectors-gemma-4-31b-it-postfix.emotion-dialogue-vectors-gemma-4-31b
Emotion vectors — gemma-4-31b (base) probed on base-generated dialogues
Data provenance (what made these activations)
Probed model: google/gemma-4-31b (base)
Input corpus: abotresol/emotion-dialogues-gemma-4-31b — two-person dialogues written by the base model (generator = probed model; 44% emotion-word leakage, documented)
Per-story pooled residual-stream activations and per-emotion mean vectors,
extracted with gemma4-emotion-vectors… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-dialogue-vectors-gemma-4-31b.emotion-selfstory-vectors-gemma-4-31b-it
Emotion vectors — gemma-4-31b-it probed on its OWN self-generated stories
Data provenance (what made these activations)
Probed model: google/gemma-4-31b-it (instruct)
Input corpus: abotresol/emotion-stories-gemma-4-31b-it — stories written by the probed model itself (generator = probed model, the reference's convention; 12 the twelve emotions, up to 256 stories each — the E6 scale corpus)
Per-story pooled residual-stream activations and per-emotion mean vectors… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-selfstory-vectors-gemma-4-31b-it.emotion-deepseek-vectors-gemma-4-31b-it
Emotion vectors from fixed-prompt DeepSeek stories
Per-emotion vectors for google/gemma-4-31b-it, built from stories written by
deepseek-v4-pro under one fixed instruction. These were the strongest
detection vectors in the project's comparison of story sources: 9 of 20 layers
cleared a bar fixed before scoring, against 5 for the model's own writing.
12 emotions, 20 layers, 5,376 dimensions per layer, from 3,070 stories.
Contents
Path
Shape
Contents… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-deepseek-vectors-gemma-4-31b-it.emotion-vectors-experiment-artifacts
Emotion-vectors replication on Gemma-4-31B: experiment artifacts
Every activation tensor, prompt set, and scored output behind the report
notebooks of gemma4-emotion-vectors
(commit a8e2352), published so replication does NOT require re-running
inference. The research record (hypotheses, pre-registered predictions,
verdicts) is the repo's TREE.md; the daily log is RESEARCH_LOG.md.
Models: google/gemma-4-31b (base) and google/gemma-4-31b-it (instruct),
bf16. Layers: range(0, 60… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-vectors-experiment-artifacts.emotion-deepseek-diverse-vectors-gemma-4-31b-it
Per-story vectors from the prompt-diversified DeepSeek corpus
Per-story vectors for google/gemma-4-31b-it, from stories written by
deepseek-v4-pro with the protagonist and setting pinned from a deterministic
8-personas by 8-settings grid.
This is the prompt-diversity condition. It tests whether forcing variety into the
prompt produces better emotion vectors than a single fixed instruction. In the
project's results it did not: at every matched sample size the fixed-prompt
corpus… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-deepseek-diverse-vectors-gemma-4-31b-it.emotion-vectors-gemma-4-31b-smoke
Emotion vectors — google/gemma-4-31b
Per-story pooled residual-stream activations and per-emotion mean vectors,
extracted with gemma4-emotion-vectors
scripts/extract_emotion_vectors.py (reference-faithful adaptation of
sinievanderben/emotion_experiment extract_emotion_vectors.py).
Corpus: snae/emotion_stories_gemma_4_4B (split train)
Layers: [0, 3, 6, 9, 12, 15, 18, 21, 24, 27, 30, 33, 36, 39, 42, 45, 48, 51, 54, 57]
Pooling: mean over non-pad tokens after position 50… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-vectors-gemma-4-31b-smoke.emotion-dialogue-vectors-gemma-4-31b-it
Emotion vectors — gemma-4-31b-it probed on its own generated dialogues
Data provenance (what made these activations)
Probed model: google/gemma-4-31b-it (instruct)
Input corpus: abotresol/emotion-dialogues-gemma-4-31b-it — two-person dialogues written by the probed model (generator = probed model; 0% leakage)
Per-story pooled residual-stream activations and per-emotion mean vectors,
extracted with gemma4-emotion-vectors
scripts/extract_emotion_vectors.py… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-dialogue-vectors-gemma-4-31b-it.