Carlosian/Qwen3.5-35B-A3B-Abliterated

Drop small internal N=16-style dev-probe numbers from method/validation; keep only real large-N eval results

Simplify degenerate-flag label to avoid misreading as real damage

Correct self-harm residual framing: incidental outcome, not deliberate design (matches Gemma card)

Update README.md

Remove paragraph conflating Guardline (child-harm/killing) with self-harm guardrail claim

Update README.md

Update README.md

Cite Nous Research CNA paper + llm-abliteration toolkit

Update model card with EvalKit benchmark results

Add independent residual-refusal audit (refusal_bench, 200 prompts)

Add files using upload-large-folder tool

initial commit