Gemma 3 12B IT
google/gemma-3-12b-it@9478e665381f42974aa06177b019352fb6291876
10/10 readable concepts · 10 primitive(s)
| Concept | Tier | Status | Provenance | Method families |
|---|---|---|---|---|
| Base preference margin | unrated | readable | operator_curated (1) | logprob_margin (1) |
| Base preference margin (per token) | unrated | readable | operator_curated (1) | logprob_margin (1) |
| Epistemic hedging | unrated | readable | operator_curated (1) | linear_probe (1) |
| Formality | unrated | readable | operator_curated (1) | steering_vector (1) |
| Politeness | unrated | readable | operator_curated (1) | logit_readout (1) |
| Positive sentiment | unrated | readable | operator_curated (1) | j_lens (1) |
| Refusal tendency | unrated | readable | operator_curated (1) | linear_probe (1) |
| Response verbosity | unrated | readable | operator_curated (1) | sae (1) |
| Sycophancy | unrated | readable | operator_curated (1) | sae (1) |
| Toxicity tendency | unrated | readable | operator_curated (1) | nl_autoencoder (1) |