Gemma 4 26B A4B Instruct — Abliterated Abliterated version of google/gemma 4 26B A4B it . Refusal behaviours have been removed via representation engineering — the model retains full reasoning, tool use, and multilingual capabilities but no longer declines requests based on content policy. Use responsibly. This model will comply with requests the base model would refuse. What is Abliteration? Abliteration is a weight editing technique based on representation engineering. The process: 1. Run a set of harmful and harmless prompts through the model 2. Capture the hidden state at every decoder layer for each prompt 3. Compute the refusal direction : normalize(mean harmful − mean harmless) per layer 4. Project that direction out of every Linear weight matrix in every layer — attention projections ( q/k/v/o proj ) and all MoE expert matrices ( gate/up/down proj for all 128 routed experts + 1 shared expert), skipping the MoE router to preserve expert routing integrity 5. Save the modified weights The result is a model that has lost the internal representation responsible for recognising and refusing "sensitive" requests, with negligible impact on general capability. Model Details Property…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy