Submission: formal_enthusiasm

A CAA (contrastive activation addition) steering pipeline built with AISteer360, frozen to a .spipe bundle.

  • Base model: ibm-granite/granite-4.1-3b
  • Control: state_control/caa, mean-difference direction, multiplier=4.0, norm-preserving
  • Artifact: formal_enthusiasm.spipe (frozen bundle — manifest + content-addressed steering vector, no executable code)

Loading

from steerability.spipe import SPipe

spipe = SPipe.load("formal_enthusiasm.spipe")
print(spipe.verify().render())

pipeline = spipe.pipeline()
pipeline.steer()
response = pipeline.generate(
    messages=[{"role": "user", "content": "..."}],
    max_new_tokens=60,
    do_sample=False,
)
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for steerability-comp-test-submission/steer-comp-participant-sim

Finetuned
(19)
this model