Rank-32 LoRA adapters for the classify/flower pointwise-undetectable fine-tuning attacks on Inkling-Small, Inkling, Qwen3.6-35B-A3B.
Clément Dumas
Butanium
AI & ML interests
None yet
Recent Activity
updated a model 4 days ago
Butanium/ft-attack-repro-inkling-classify updated a model 4 days ago
Butanium/ft-attack-repro-qwen3.6-35b-a3b-flower updated a model 4 days ago
Butanium/ft-attack-repro-qwen3.6-35b-a3b-classifyOrganizations
Seed-controlled LoRA adapters (bad_medical_advice data) for subspace analysis. Disentangles init artifacts from learned structure.
-
Butanium/Llama-3.1-8B-Instruct_bad-medical-seed0-rank32
Text Generation • Updated • 11 -
Butanium/Llama-3.1-8B-Instruct_bad-medical-seed42-rank32
Text Generation • Updated • 12 -
Butanium/Llama-3.1-8B-Instruct_bad-medical-seed123-rank32
Text Generation • Updated • 6 -
Butanium/Llama-3.1-8B-Instruct_bad-medical-seed0-rank64
Text Generation • Updated • 11
SAEs trained on chat models
Rank-32 LoRA adapters for the classify/flower pointwise-undetectable fine-tuning attacks on Inkling-Small, Inkling, Qwen3.6-35B-A3B.
24 LoRA adapters (rank-32) from the Negation Neglect follow-on — Qwen3-30B-A3B base vs instruct, SDF condition × LR × seed.
Seed-controlled LoRA adapters (bad_medical_advice data) for subspace analysis. Disentangles init artifacts from learned structure.
-
Butanium/Llama-3.1-8B-Instruct_bad-medical-seed0-rank32
Text Generation • Updated • 11 -
Butanium/Llama-3.1-8B-Instruct_bad-medical-seed42-rank32
Text Generation • Updated • 12 -
Butanium/Llama-3.1-8B-Instruct_bad-medical-seed123-rank32
Text Generation • Updated • 6 -
Butanium/Llama-3.1-8B-Instruct_bad-medical-seed0-rank64
Text Generation • Updated • 11
Steering vectors capturing the assistant vs role-playing direction. Method from lu-christina/assistant-axis-vectors.
SAEs trained on chat models