Jacobian lens for Qwen2.5-0.5B-Instruct
A fitted Jacobian lens (lens.pt, 23 source
layers, d_model 896) for Qwen/Qwen2.5-0.5B-Instruct, fitted on 200 wikitext-103 chunks of
128 tokens.
Used by the silent-thoughts workbench and its demo Space.
The workbench also reads the model's committed decision before it speaks — a pre-emission prompt-injection monitor — and can rewrite it. That read→write loop is dual-use: the same machinery that catches an injection is a silent censorship tool. See DANGERS.md.
import jlens
lens = jlens.JacobianLens.from_pretrained("MildHotSauce/jlens-qwen2.5-0.5b-instruct")
Fitted by Dave Ralston — [email protected]
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support