A fine-tuned module that enables an LLM to generate outputs directly from non-sequential cache inputs.