“trl”的搜索结果

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

中文简介正在准备中。当前显示原文说明。

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

中文简介正在准备中。当前显示原文说明。

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.

中文简介正在准备中。当前显示原文说明。

組織の生きているペイン（secretary / dispatcher / worker）へ tmux で**直接入る** ための **コマンドを出力するだけ** の read-only スキル。`mcp__org-broker__list_panes`（論理ペイン） と `/usr/bin/tmux -L claude-org-broker list-panes -a`（pane_id ↔ session マッピング）を pane_id（%N）で突き合わせ、各ペインに role + name(task_id) ラベル付きの attach コマンドを 生成する（既定は読取書込 = `-r` 無し / 読取専用 `-r` はオプション / デタッチは `Ctrl-b d`）。 外側が tmux のときに今の tmux を分割してワーカー（と dispatcher）を並べる `split-window` コマンドも出す。 特定のペインの画面を直接見たい・入りたい・並べて見たい、attach コマンドを知りたいという依頼で発動。 自分で attach / split はせず・ペインも一切変更しない（コマンド文字列とペイン一覧表を表示するのみ）。 作業委譲（org-delegate）・ダッシュボードで状況一覧（org-dashboard）・watcher 停止（org-attention-stop） には発動しない。

中文简介正在准备中。当前显示原文说明。

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

中文简介正在准备中。当前显示原文说明。

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。

Expert knowledge of Folketrygdloven (National Insurance Act) processing in melosys-api. Use when: (1) Understanding FTRL bestemmelser (§2-5 to §2-13), (2) Debugging yrkesaktiv/ikke-yrkesaktiv/pensjonist processing flows, (3) Understanding trygdedekning options and combinations, (4) Investigating vilkår for each bestemmelse, (5) Understanding medlemskapsperiode generation logic, (6) Debugging frivillig vs pliktig membership determination.

中文简介正在准备中。当前显示原文说明。

查看并操作tmux中运行的交互式命令行程序，列出会话和分屏窗格，获取当前输出或历史记录，发送文字与按键，并监测输入提示或处理完成。

• 查看交互式程序的输出
• 向运行中的会话发送文字和按键
• 监测输入等待或处理完成

帮助您使用Unsloth，以LoRA和QLoRA微调语言模型，探索减少GPU内存占用的训练方式，并依据官方文档整理的参考资料查询功能与API、实现代码和排查错误。

• 开始使用LoRA或QLoRA微调
• 查询Unsloth功能与API
• 排查训练代码中的错误

将功能开发、重构和PR审查交给OpenCode CLI，跟踪长时间任务的进度、恢复会话，并汇报修改文件和测试结果，也支持在独立目录中并行执行。

• 委派功能开发和重构
• 审查PR中的缺陷与测试遗漏
• 监控并恢复长时间开发会话

Prepare and review LoRA or QLoRA settings for adapting language models using example responses. Choose trainable layers, adapter capacity, learning rate, and memory-related options, then produce configuration values for a downstream training script.

中文简介正在准备中。当前显示英语简介。

• Preparing a LoRA configuration for supervised fine-tuning
• Reviewing rank, alpha, and learning rate settings
• Choosing LoRA or QLoRA based on GPU memory

Choose a model training method and configuration from paired response preferences or individual positive and negative ratings. Covers DPO, ORPO, KTO, and SimPO, including pair construction, repeated training rounds, and troubleshooting learning settings.

中文简介正在准备中。当前显示英语简介。

• Choose a training method for paired response preferences
• Select a method for individual positive and negative ratings
• Review learning settings when DPO training loses general capability

Set up machine learning training and inference on NVIDIA DGX Spark. Get guidance on choosing containers and packages, resolving CUDA compatibility errors, and checking that PyTorch can access the GPU before starting a workload.

中文简介正在准备中。当前显示英语简介。

• Setting up a new DGX Spark for training or inference
• Diagnosing CUDA compatibility and import errors
• Choosing between containers and direct pip installation

Choose the right WeKnora command to answer questions from a knowledge base or retrieve source passages. Covers cited answers, custom agents, document searches, and continuing a conversation through the command-line tool.

中文简介正在准备中。当前显示英语简介。

• Ask questions with citations from a knowledge base
• Retrieve source passages for further analysis
• Find documents by title or filename

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

Get guidance on adapting LLMs with Unsloth, including implementation, API usage, and debugging. Covers LoRA and QLoRA, methods for adjusting models with fewer trainable parameters, with a focus on training speed and memory use.

中文简介正在准备中。当前显示英语简介。

• Learning how to fine-tune models with Unsloth
• Exploring LoRA and QLoRA for efficient training
• Implementing features using Unsloth APIs

Fine-tune large language models with LoRA and QLoRA while limiting GPU memory use. Guides training settings and adapter management, using small sets of task-specific weights that can be saved, switched, or merged into a base model.

中文简介正在准备中。当前显示英语简介。

• Fine-tuning an LLM with limited GPU memory
• Choosing LoRA settings and training layers
• Switching task-specific adapters on one base model

Create and run tests for interactive command-line tools and terminal interfaces using @microsoft/tui-test. Simulate keyboard input, check visible text, and compare screen snapshots in a separate terminal environment for each test.

中文简介正在准备中。当前显示英语简介。

• Testing selection prompts with arrow keys and Enter
• Checking terminal screen snapshots
• Automating text entry and keyboard navigation

Add and troubleshoot the json-render inspection panel in SolidJS applications. This skill covers panel setup for examining specifications, editing state, reviewing action and stream timelines, browsing component catalogs, and selecting page elements for inspection.

中文简介正在准备中。当前显示英语简介。

• Add an inspection panel to a SolidJS app
• Inspect JSON specifications and edit application state
• Review action and stream timelines

Populate a running CARLA simulation with autonomous traffic, evenly spaced vehicles in one lane, or a single vehicle for control and sensors. Filter vehicle models, keep automated driving active, and remove vehicles when the scene is finished.

中文简介正在准备中。当前显示英语简介。

• Fill a map with autonomous traffic
• Place vehicles at fixed intervals in one lane
• Create an ego vehicle for control and sensors

Add pedestrians to a CARLA map and let them walk autonomously. Adjust crowd size, walking speed, road-crossing frequency, and placement seeds, create stationary crowds, and remove pedestrians together with their movement controllers in the correct order.

中文简介正在准备中。当前显示英语简介。

• Adding pedestrians who walk around a map
• Reusing pedestrian placement with a fixed seed
• Adjusting walking speeds and road crossings

Get guidance on fine-tuning language models with Unsloth, including LoRA and QLoRA methods for training with fewer resources. The skill points to documentation for understanding APIs, implementing training code, debugging problems, and exploring speed and memory optimizations.

中文简介正在准备中。当前显示英语简介。

• Exploring faster training with lower memory use
• Learning about LoRA and QLoRA fine-tuning
• Checking Unsloth features and APIs

Adapt large language models to specific tasks through fine-tuning, or additional training. Prepares and validates datasets, configures LoRA or QLoRA to train a small subset of parameters, and supplies training settings, evaluation scripts, and deployment optimization guidance.

中文简介正在准备中。当前显示英语简介。

• Preparing and validating JSONL training data
• Choosing LoRA or QLoRA for available GPU memory
• Configuring training and monitoring overfitting

Design and implement command-line tools with subcommands, options, interactive prompts, and progress indicators. Get command structures, configuration handling, clear error messages, and shell completion scripts, with guidance for automated use and testing across operating systems.

中文简介正在准备中。当前显示英语简介。

• Build a command-line tool with subcommands and options
• Add interactive prompts and progress indicators
• Provide bash, zsh, and fish command completion