Skip to content
#

decoder-attribution

Here is 1 public repository matching this topic...

How much of an LLM agent’s apparent behavior comes from the decoder rather than the model alone? SamplerScope holds the model’s logits fixed, changes only the decoding method, and measures how this alters outcomes in small decision environments.

  • Updated Aug 17, 2026
  • Python

Add this topic to your repo

To associate your repository with the decoder-attribution topic, visit your repo's landing page and select "manage topics."

Learn more