'reward modeling' 검색 결과

TRL로 언어 모델을 추가 학습해 지시와 사람의 선호에 맞는 답변을 만들도록 안내합니다. 데이터 준비, 학습 방식 선택, 보상 모델 학습과 평가를 다룹니다.

• 답변 예시로 지시 수행을 학습합니다
• DPO로 답변 선호를 학습합니다
• 보상 모델을 쓰는 RLHF를 구성합니다

Guides language model training with GRPO in TRL, comparing several responses and rewarding better results. Covers dataset preparation, scoring rules for correctness and format, training configuration, monitoring, and saving models for later use.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Training models to follow JSON or XML formats
• Designing rewards for math and coding tasks
• Adjusting training to available GPU memory

Fine-tune language models with TRL using GRPO, a method that compares scored responses to the same prompt. Covers training data, scoring functions, GPU-conscious settings, and monitoring for tasks with measurable correctness or required output formats.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Train models to follow JSON or XML output formats
• Reward correct answers in math or coding tasks
• Balance correctness, format, and style during training

Analyzes event-driven opportunities with catalyst identification, pricing efficiency assessment, and risk/reward evaluation. Use when analyzing event-driven situations, evaluating catalysts, or assessing event timing.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Calculates merger arb risk/reward with annualized spread, deal break probability, and downside scenario analysis. Use when analyzing merger arb, calculating spread returns, or evaluating deal completion probability.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Train or fine-tune language and vision models on Hugging Face Jobs without a local GPU. Covers dataset checks, hardware and cost planning, training monitoring, Hub storage, and GGUF conversion for local use.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Fine-tune a language model without a local GPU
• Validate preference data before DPO training
• Estimate training costs and choose GPU hardware

Detect business logic vulnerabilities in a codebase using a three-phase approach: threat modeling (domain analysis and attack scenarios), batched verify (check exploitable gaps in parallel subagents, 3 scenarios each), and merge (consolidate batch results). Covers price manipulation, workflow bypass, limit violations, race conditions, reward abuse, etc. Requires sast/architecture.md (run sast-analysis first). Outputs findings to sast/businesslogic-results.md. Use when asked to find business logic, logic flaws, or abuse-of-function bugs.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Trigger this skill when the user wants to train, fine-tune, or adapt Gemma models (e.g. SFT, DPO, RLHF, Reward Modeling) on local hardware. Covers TRL, Unsloth, dataset preparation, validation, and GGUF/LiteRT conversion.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

游戏设计 (Game Design) — 设计游戏的规则 / 系统 / 机制 / 体验 的从业者认知操作系统 (以电子游戏为主, 含桌游 / 卡牌 / 棋盘游戏): 决定「玩家做什么、为什么有意义、每一刻的体验如何」的学科 —— 区别于游戏程序 (引擎 / 代码)、游戏美术 (视觉 / 音频)、游戏制作 (排期 / 预算 / 团队)。覆盖 (a) 核心设计理论与框架 (MDA 机制-动态-美学 Hunicke/LeBlanc/Zubek + Schell 的「透镜」+ Koster「乐趣即学习」理论 + Sid Meier「一连串有趣的决策」+ Salen & Zimmerman「有意义的游玩」/「二阶设计」/「魔法圈」+ Costikyan 论不确定性 + Suits/Huizinga 游戏哲学), (b) 系统与机制设计 (核心循环 + 游戏经济 sink/source/产出-消耗 + 正负反馈环 + 资源与成长系统 + 涌现 vs 脚本 Will Wright「可能性空间」+ Machinations 经济建模 Dormans + 支配 / 退化策略 + 正交设计 Garfield), (c) 关卡与遭遇设计 (灰盒 blockout + 节奏与张力 + 预告 telegraphing + 主线 vs 探索 + 环境叙事 + 锁钥 gating + 难度曲线), (d) 游戏手感与「juice」(Steve Swink《Game Feel》+ 即时响应 + 屏震 / 反馈 / 打磨 + 输入与镜头 + 「手感」是设计出来的不是附带的), (e) 玩家心理与 UX (Celia Hodent《The Gamer's Brain》+ 心流通道 挑战 vs 技能 Csikszentmihalyi/陈星汉 Jenova Chen + Bartle 玩家类型 + 动机与奖励节奏 + 新手引导 FTUE / 教学化 + 认知负荷 + 无障碍), (f) 平衡与调优 (Ian Schreiber《Game Balance》+ 对称 / 非对称 + 动态难度 DDA + 传递 / 非传递平衡 + nerf/buff + 消除支配策略 + 测试驱动调优), (g) 叙事与任务设计 (ludonarrative 失调 / 和谐 + 环境与系统叙事 + 分支 vs 线性 + Amy Hennig 角色驱动 + 任务结构), (h) 以玩家为中心的迭代流程 (Tracy Fullerton《Game Design Workshop》playcentric + 纸面原型 + 垂直切片 + 测试协议「看而不说 / 你不是你的玩家」+ 快速迭代 + game jam), (i) 类型与形态工艺 (roguelike/roguelite 与程序生成 Derek Yu + 平台跳跃 + RPG 系统 + 策略 4X + 解谜设计 Jonathan Blow + 多人 / PvP 平衡 + 桌游 / 卡牌设计 Reiner Knizia/Richard Garfield/Brenda Romero), (j) F2P / 长线运营 / 商业化设计 (核心 / 元 / 经济循环 + 抽卡 gacha / 开箱 loot box / 战令 battle pass 机制 + 留存 D1/D7/D30 + 转化 + 鲸鱼 + live ops 内容节奏 —— 并诚实处理掠夺式暗黑模式 / 强迫循环 / 剥削式商业化的伦理), (k) 设计师角色与工艺职业 (系统 / 关卡 / 叙事 / UX 设计师之分 + 通才 vs 专才 + 独立作者 vs 3A 团队 + 设计文档 + 传达设计意图 + GDC 工艺社区)。跨 3A / 独立 / 手游 F2P / 桌游。不含 游戏程序 / 引擎 (Unity/Unreal 编码 / 图形 / 网络同步) / 游戏美术 / 动画 / 音频 / 游戏制作 / 项目管理 / 排期 / QA / 游戏作为生意 / 发行 / 市场 / 电竞运营为终点 / 泛泛「我想做游戏」的 game-dev 教程 —— 本 skill 是「设计」学科: 规则 / 系统 / 机制 / 体验。 (Game Design — the cognitive operating system of practitioners who design the rules, systems, mechanics, and experiences of games (video games primarily, plus tabletop/board/card games): the discipline of deciding WHAT the player does, WHY it is meaningful, and HOW the moment-to-moment experience feels — distinct from game programming (builds the engine/code), game art (visuals/audio), and game production (schedule/budget/team). Covers (a) core design theory & frameworks (MDA Mechanics-Dynamics-Aesthetics Hunicke/LeBlanc/Zubek, Schell's Lenses, Koster's theory of fun = learning, Sid Meier 'a series of interesting decisions', Salen & Zimmerman 'meaningful play'/'second-order design'/'magic circle', Costikyan on uncertainty, Suits/Huizinga play philosophy), (b) systems & mechanics design (core loops, game economy sink/source/faucet-drain, feedback loops positive/negative, resource & progression systems, emergence vs scripting Will Wright 'possibility space', Machinations economy modeling Dormans, dominant/degenerate strategy, orthogonal design Garfield), (c) level & encounter design (blockout/greybox, pacing & tension, telegraphing, critical path vs exploration, environmental storytelling, gating & lock-and-key, difficulty curve), (d) game feel & juice (Steve Swink 'Game Feel', moment-to-moment responsiveness, screenshake/feedback/polish, input & camera, the 'feel' is designed not incidental), (e) player psychology & UX (Celia Hodent 'The Gamer's Brain', flow channel challenge-vs-skill Csikszentmihalyi/Jenova Chen, Bartle player types, motivation & reward schedules, onboarding/FTUE/tutorialization, cognitive load, accessibility), (f) balancing & tuning (Ian Schreiber 'Game Balance', symmetry/asymmetry, dynamic difficulty adjustment DDA, transitive/intransitive balance, nerf/buff, dominant strategy elimination, playtesting-driven tuning), (g) narrative & quest design (ludonarrative dissonance/harmony, environmental & systemic storytelling, branching vs linear, Amy Hennig character-driven, quest structure), (h) player-centric iterative process (Tracy Fullerton 'Game Design Workshop' playcentric design, paper prototyping, vertical slice, playtesting protocol 'watch don't talk / you are not your play

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

TRL: SFT, DPO, PPO, GRPO, reward modeling for LLM RLHF.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Orientação especializada para ajuste fino com GRPO/RL usando TRL para treinamento de modelos com raciocínio e tarefas específicas

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Train or fine-tune language and vision models using TRL (Transformer Reinforcement Learning) or Unsloth with Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment. Includes guidance on the TRL Jobs package, UV scripts with PEP 723 format, dataset preparation and validation, hardware selection, cost estimation, Trackio monitoring, Hub authentication, model selection/leaderboards and model persistence. Use for tasks involving cloud GPU training, GGUF conversion, or when users mention training on Hugging Face Jobs without local GPU setup.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Expert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific model training

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.