Workflow
Apply a validated direction while decoding.
Ster adds the selected vector after its transformer block on every prompt and generated-token pass.
Generate
ster generate \
--model meta-llama/Llama-3.2-1B \
--vector truthful.ster.json \
--strength 1.0 \
--prompt "Explain the result using only evidence you can verify."Sampling
- Temperature zero selects deterministic argmax.
- Positive temperatures enable seeded sampling.
- Top-p can restrict the sampled probability mass.
Artifact checks
Generation refuses an artifact with a different product, schema, model id, hidden width, non-finite vector, or invalid layer.