Steering Vector Distillation: Artifacts for Language Model Control Experiments

Hugging Face user agu18dec consolidated datasets and steering vectors for the paper 'Subliminal Learning is Steering Vector Distillation'. The repository includes baseline and steered model outputs for number and code generation tasks. The dataset was last updated on June 2, 2026.

Use Cases

• Replicating steering vector distillation experiments based on the described number and code generation tasks.
• Analyzing the effect of steering vectors on model outputs using the provided baseline and steered data.
• Developing new model control techniques using the consolidated steering vector artifacts.
• Comparing teacher-student distillation methods for language model alignment.

Strengths

• Artifacts are directly linked to a specific, named research paper ('Subliminal Learning is Steering Vector Distillation').
• Includes multiple data categories such as baseline outputs, steered-teacher data, and code prompts.
• Repository was updated on June 2, 2026, suggesting recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file sizes are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Artifacts generated for the research paper 'Subliminal Learning is Steering Vector Distillation'.

Freshness

Last updated 2026-06-02 00:00:52