メニュー
このリポジトリの skills
SkillsMP は TheAdaply/re-forge から 122 件の skill を収集しています。skill を開くとソースと詳細を確認できます。
TheAdaply/re-forge収集済み skill 122 件中 40 件を表示しています。
Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you want faster inference than vLLM on workloads with heavy prefix sharing. The…
原文の言語: 英語
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…
原文の言語: 英語
Multi-cloud orchestration for ML workloads with automatic cost optimization. Use when you need to run training or batch jobs across multiple clouds, leverage spot instances with auto-recovery, or optimize GPU costs across providers.
原文の言語: 英語
Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.
原文の言語: 英語
State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.
原文の言語: 英語
Provides guidance for experiment tracking with SwanLab. Use when you need open-source run tracking, local or self-hosted dashboards, and lightweight media logging for ML workflows.
原文の言語: 英語
Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's ML visualization toolkit
原文の言語: 英語
Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency. Use for production deployment on NVIDIA GPUs (A100/H100), when you need 10-100x faster inference than PyTorch, or for serving models with quantization (FP8/INT4),…
原文の言語: 英語
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
原文の言語: 英語
Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing…
原文の言語: 英語
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…
原文の言語: 英語
Pointers for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization
原文の言語: 英語
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
原文の言語: 英語
Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints,…
原文の言語: 英語
Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform
原文の言語: 英語
Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for…
原文の言語: 英語
Evaluates code generation models across HumanEval, MBPP, MultiPL-E, and 15+ benchmarks with pass@k metrics. Use when benchmarking code models, comparing coding abilities, testing multi-language support, or measuring code generation quality. Industry standard…
原文の言語: 英語
Dispatch the Engineering Team to implement something. Takes a task description (or a research SYNTHESIS reference) and runs the full plan-then-build protocol with verifier + reviewer loop. Use this skill IMMEDIATELY when the user asks to implement, build,…
原文の言語: 英語
Dispatch the Research Team on any question. Amplifies a terse prompt into a full multi-specialist investigation with adversarial gates (skeptic + adversary + moderator + evaluator). Use this skill IMMEDIATELY when the user asks to research something,…
原文の言語: 英語
Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms. Use when you want clean RL abstractions, easy algorithm experimentation, or scalable training with Monarch and TorchTitan.
原文の言語: 英語
Use this skill to compare coding agents (re-forge agents, Claude Code, Aider, Codex, etc.) head-to-head on reproducible tasks, measuring pass rate, cost, time, and consistency instead of relying on vibes. Activate when choosing between agents/models,…
原文の言語: 英語
Use this skill when an agent run is failing repeatedly, burning tokens without progress, looping on the same tools, or drifting from the task. It is a structured self-debugging workflow — capture the failure, diagnose the root cause, apply contained recovery,…
原文の言語: 英語
Use this skill for regression testing in AI-assisted development, where the same model writes AND reviews code and so carries identical blind spots into both steps. It covers sandbox/mock-mode API testing without a database, an automated bug-check workflow,…
原文の言語: 英語
Use this skill when the user asks to make something faster, try many variants, run recursive optimization, benchmark latency/throughput/cost, or choose the best implementation by repeated measured tests. It converts ambitious targets like "make it 20x faster"…
原文の言語: 英語
Use this skill to measure performance baselines and detect regressions in re-forge — before/after a PR, when users report "it feels slow", or before a launch when you must meet performance targets. It covers page (Core Web Vitals), API (latency percentiles),…
原文の言語: 英語
Use this skill when joining a new project or opening an unfamiliar repo in re-forge for the first time. Analyzes the codebase and produces a structured onboarding guide (architecture map, entry points, conventions) plus a starter project instructions file.…
原文の言語: 英語
Use this skill to produce thorough, cited research reports synthesized from multiple web sources. Activate when the user wants to research a topic in depth, do competitive analysis, evaluate a technology, size a market, or run due diligence — anytime an…
原文の言語: 英語
Use this skill in re-forge when designing error types, adding retries/circuit breakers for unreliable dependencies, reviewing API endpoints for missing error handling, building user-facing error messages, or debugging silent failures and cascading errors.…
原文の言語: 英語
Use this skill to set up eval-driven development (EDD) for re-forge — treating evals as the "unit tests of AI development." Activate when defining pass/fail criteria for an agent task, measuring reliability with pass@k metrics, building regression suites for…
原文の言語: 英語
Dispatch the Evolution Team — re-forge's self-evolving watch team whose sole job is to keep the workforce ahead of a fast-moving ecosystem. Use this skill when the user asks to "check what's new in Claude Code / Cursor / Codex", "scan for new agent features",…
原文の言語: 英語
Use this skill in re-forge when choosing a branching strategy, writing commit messages or PR descriptions, deciding merge vs rebase, resolving conflicts, or managing releases and tags. Covers conventional commits, code-review checklists, and collaborative…
原文の言語: 英語
Use this skill whenever you add authentication or authorization, handle user input or file uploads, work with secrets or credentials, create API endpoints, implement payment or blockchain features, or store/transmit sensitive data in re-forge. It provides a…
原文の言語: 英語
Use this skill to audit re-forge's agent configuration (the `.claude/` directory — CLAUDE.md, settings.json, MCP configs, hooks, and agent/skill definitions) for security vulnerabilities, misconfigurations, and prompt-injection risks. Activate when setting up…
原文の言語: 英語
Use this skill to decide WHEN to manually compact context during long re-forge sessions, instead of letting arbitrary auto-compaction interrupt work mid-task. Activate on long multi-phase sessions (research → plan → implement → test), when switching between…
原文の言語: 英語
Use this skill whenever you write a new feature, fix a bug, or refactor code in re-forge. It enforces strict test-driven development — tests BEFORE code, a verified RED→GREEN→REFACTOR cycle, and 80%+ coverage across unit, integration, and E2E tests. Activate…
原文の言語: 英語
Use this skill to run a comprehensive quality gate after completing a feature, before opening a PR, or after refactoring in re-forge. It walks build → types → lint → tests → security scan → diff review and emits a single READY / NOT READY verification report.…
原文の言語: 英語
Dispatch the Security & Review Team on any security task. Auto-detects language, dependency manager, and existing security tooling. Use this skill IMMEDIATELY when the user asks for a security audit, dependency CVE check, secrets scan, threat model, license…
原文の言語: 英語
Dispatch the Documentation & Knowledge Team on any documentation task. Auto-detects project language, framework, and existing docs. Use this skill IMMEDIATELY when the user asks to generate docs, update README, write API docs, create changelogs, document…
原文の言語: 英語
Dispatch the Testing & QA Team on any testing task. Auto-detects language, test framework, coverage tool, and existing test patterns. Use this skill IMMEDIATELY when the user asks to write tests, improve coverage, run mutation testing, detect regressions,…
原文の言語: 英語
Dispatch the Capability Forge to detect workforce gaps, scout for existing implementations, or author new skills/MCPs/plugins. Use this skill when the user asks "what capability are we missing", "build a skill for X", "check MCP Registry for Y", "what skills…
原文の言語: 英語