abotresol/emotion-combined-trajectories-gemma-4-31b
emotion-combined-trajectories-gemma-4-31b BASE-model condition of the per-token trajectory stories (companion to emotion-combined-trajectories-gemma-4-31b-it, same corpus and layout). One shard per story from a teacher-forced forward pass through google/gemma-4-31b (base, bf16, right padding), residual captured at layers [6, 15, 24, 33, 42, 51]. Provenance caveats specific to this condition The stories were generated by the INSTRUCT model (base cannot chat-write… See the full description on the dataset page: https://huggingface.co/datasets/abotresol/emotion-combined-trajectories-gemma-4-31b.
emotion-combined-trajectories-gemma-4-31b
BASE-model condition of the per-token trajectory stories (companion to emotion-combined-trajectories-gemma-4-31b-it, same corpus and layout). One shard per story from a teacher-forced forward pass through google/gemma-4-31b (base, bf16, right padding), residual captured at layers [6, 15, 24, 33, 42, 51].
Provenance caveats specific to this condition
- The stories were generated by the INSTRUCT model (base cannot chat-write stories), so for this condition generator != probed model. Documented, unavoidable.
- Probes: 171 base-external-corpus contrasts + 24 fixed-seed random null directions (195 total,
probe_labels.json). There is NO self-generated story source for base, for the same reason.
Schema, centered-cosine recipe, and precision notes: see the -it companion dataset card. Extraction: 5,888 stories, 0 errors (extract.log).
