artifact
active
artifact:deepseek-r1-saes-general-reasoning-and-mathDeepSeek-R1 SAEs (general reasoning and math)
Two open-sourced SAEs, the first public interpreters for a true reasoning model at this scale.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Under the Hood of a Reasoning Model - Goodfireintroducesmentions
Datasets (3)
dataset
- Custom reasoning dataset on whose R1 activations the general SAE was trained.
- OpenR1-MathusesLarge mathematical reasoning dataset used for the math SAE.
- Activations from the undistilled R1 on which SAEs were trained.
Concepts (1)
concept
- Reasoning modelaboutModels like R1 that produce chain-of-thought and backtrack; claimed qualitatively different from non-reasoning LMs.
Institutes (1)
institute
- GoodfirecitesAI research company; authors' affiliation; develops tools including EVEE and publishes research on genomic foundation models.