magiccodingman/QwQ-32B-abliterated-131k-GGUF-Yarn-Imatrix

该数据集为QwQ-32B-Abliterated模型构建过程中使用的数据集，包含了大约1600万个标记，经过过滤以消除拒绝性回应，如“Im sorry”。此外，还包括约30万个自定义编写的提示，用于增强JSON结构稳定性、改进多智能体编排的连续性以及强化背景、语调约束和世界一致性行为。The dataset used in the construction of the QwQ-32B-Abliterated model consists of approximately 16 million tokens, filtered to eliminate refusal responses such as Im sorry. In addition, it includes about 300,000 custom-authored prompts designed to enhance JSON structure stability, improve continuity in multi-agent orchestration, and reinforce subtext, tone restraint, and world-consistent behavior.