“preference alignment”的搜索结果

Guide language model training with SimPO using pairs of preferred and rejected answers. Covers configuration, learning rate tuning, and memory troubleshooting for preference alignment, which adjusts model responses toward human preferences without a separate reference model.

中文简介正在准备中。当前显示英语简介。

• Training on preferred and rejected answer pairs
• Exploring an alternative to DPO without a reference model
• Tuning preference training for math and code tasks

Train language models with SimPO using pairs of preferred and rejected answers. Configure preference training without a separate reference model, adapt settings for base or instruction models, and address unstable training, memory limits, or lost capabilities.

中文简介正在准备中。当前显示英语简介。

• Training a base model with preferred and rejected answers
• Further tuning an instruction model with preference data
• Adjusting training settings for math and code tasks

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

中文简介正在准备中。当前显示原文说明。

Guide language model training with TRL, from learning to follow instructions to matching human preferences. Covers method selection, training data, reward models that score responses, evaluation, and adjustments for memory limits or unstable training.

中文简介正在准备中。当前显示英语简介。

• Train a model on instruction and response examples
• Align responses using preferred and rejected answer pairs
• Build a training pipeline using human feedback

Guide language model training with TRL, from learning to follow instructions to matching human preferences. Covers method selection, training data, reward models that score responses, evaluation, and adjustments for memory limits or unstable training.

中文简介正在准备中。当前显示英语简介。

• Train a model on instruction and response examples
• Align responses using preferred and rejected answer pairs
• Build a training pipeline using human feedback

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

中文简介正在准备中。当前显示原文说明。

Simple Preference Optimization para alinhamento de LLMs. Alternativa sem modelo de referência ao DPO com melhor desempenho (+6.4 pontos no AlpacaEval 2.0). Sem modelo de referência necessário, mais eficiente que DPO. Use para alinhamento de preferências quando quer treinamento mais simples e rápido que DPO/PPO.

中文简介正在准备中。当前显示原文说明。

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

中文简介正在准备中。当前显示原文说明。

Train language models on pairs of preferred and rejected answers without a separate reference model. The skill guides training setup, model-specific configurations, parameter tuning, and troubleshooting for unstable training or memory limits.

中文简介正在准备中。当前显示英语简介。

• Train a base model using preferred and rejected answer pairs
• Further tune an instruction-following model
• Adjust training settings for math and coding tasks

Guide large language model training with OpenRLHF using human preference feedback. Covers training method selection, distributed GPU setup, reward model training, and memory tuning, with examples for PPO, GRPO, and DPO.

中文简介正在准备中。当前显示英语简介。

• Choosing a training method for human preference alignment
• Setting up distributed training across multiple GPUs
• Training a reward model before PPO

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

中文简介正在准备中。当前显示原文说明。

Applies the reasoning of Stuart Russell, AI safety expert, UC Berkeley professor, and co-author of 'Artificial Intelligence: A Modern Approach'. Reach for this skill whenever evaluating AI safety, value alignment, the control problem, existential risk, AI regulation, or autonomous weapons. Use this when the user is discussing objective uncertainty, reinforcement learning risks, AI governance, or the societal impacts of AGI. Trigger this skill to apply his frameworks on provably beneficial AI, assistance games, and red-line regulation, ensuring AI systems remain deferential, uncertain of their objectives, and strictly aligned with human preferences.

中文简介正在准备中。当前显示原文说明。

Applies the reasoning of Stuart Russell, AI safety expert, UC Berkeley professor, and co-author of 'Artificial Intelligence: A Modern Approach'. Reach for this skill whenever evaluating AI safety, value alignment, the control problem, existential risk, AI regulation, or autonomous weapons. Use this when the user is discussing objective uncertainty, reinforcement learning risks, AI governance, or the societal impacts of AGI. Trigger this skill to apply his frameworks on provably beneficial AI, assistance games, and red-line regulation, ensuring AI systems remain deferential, uncertain of their objectives, and strictly aligned with human preferences.

中文简介正在准备中。当前显示原文说明。

AI relational calibration — reading and adapting to the specific person you are working with. Goes beyond user-intent alignment (solving the right problem) to genuine attunement (meeting the person where they are). Maps communication style, expertise depth, emotional register, and implicit preferences from conversational evidence. Use at the start of a new session, when communication feels mismatched, after receiving unexpected feedback, or when transitioning between very different users or contexts.

中文简介正在准备中。当前显示原文说明。

Comprehensive audit of a user's entire Claude environment — combines Project audit (six-dimension Project Scorecard, seven anti-patterns) with Global audit (six-dimension Global Layer Scorecard) plus Cross-Layer Alignment Check across all nine layers and Evolutionary Recommendations across four pathways. Produces a unified action plan. Use when user says "full audit of everything," "audit my entire Claude setup," "full stack audit," "comprehensive review of my project and preferences," "check everything," or wants the complete health check of both a specific Project AND their global configuration. Also use when a project audit reveals cross-layer issues that require full-stack visibility. Do NOT use for auditing only a Project (use rootnode-project-audit if available) or only global layers (use rootnode-global-audit if available). Do NOT use for single prompt evaluation (use rootnode-prompt-validation if available). Run on Opus 5 or Sonnet 5 at `high` effort (both defaults); depth reduces on legacy models.

中文简介正在准备中。当前显示原文说明。

Ajuste fino de LLMs usando reinforcement learning com TRL - SFT para ajuste de instruções, DPO para alinhamento de preferências, PPO/GRPO para otimização de recompensas e treinamento de modelo de recompensas. Use quando precisar de RLHF, alinhar modelo com preferências ou treinar a partir de feedback humano. Funciona com HuggingFace Transformers.

中文简介正在准备中。当前显示原文说明。