pytorch-kernels
Skill17 starsChanged 8 months agoNo licenceView on GitHub
Custom and community GPU kernels for PyTorch transformer training and inference. Use when optimizing attention (FlashAttention, FlexAttention, SDPA, NATTEN, sliding window, paged, neighborhood), MLP fusion (Liger, SwiGLU, GeGLU, FusedMLP), normalization (RMSNorm, LayerNorm), RoPE, CrossEntropy, or optimizers (fused AdamW). Especially relevant for DiT, MMDiT, video DiT, S3DiT architectures. Also covers Colab A100 setup for flash-attn, torch.compile kernel fusion, and Triton kernel libraries like Unsloth and Liger Kernel.
In this repository49 other files it gives its agents.
- animejs-reactSkill
- apple-releasesSkill
- b2-native-pythonSkill
- clerk-typescriptSkill
- dockerSkill
- electron-appsSkill
- electron-macSkill
- fira-sans-typographySkill
- gemini-api-javascriptSkill
- gmail-apiSkill
- google-calendar-apiSkill
- google-calendar-typescriptSkill
- grok-api-sdkSkill
- huggingface-datasetsSkill
- huggingface-datasetsSkill
- imagededup-pythonSkill
- inspect-ai-pythonSkill
- iosSkill
- lancedb-typescriptSkill
- lmstudio-pythonSkill
- lmstudio-typescriptSkill
- macos-audio-captureSkill
- macosSkill
- mcp-client-typescriptSkill
- mcp-server-typescriptSkill
- mdxeditor-reactSkill
- milkdown-editorSkill
- modal-comSkill
- nango-integrationsSkill
- nextjs-docsSkill
- pydantic-ai-pythonSkill
- pydantic-pythonSkill
- python-reviewSkill
- pytorch-perf-trainingSkill
- runpod-pods-pythonSkill
- shared-ci-cdSkill
- slack-agents-typescriptSkill
- solid-swiftSkill
- space-grotesk-typographySkill
- swift-coreSkill
- swiftui-proSkill
- tailscale-networkingSkill
- tauri-appsSkill
- telegrambot-typescriptSkill
- typescript-code-qualitySkill
- vast-ai-pythonSkill
- vercel-ai-sdkSkill
- websockets-fastapiSkill
- workos-auth-typescriptSkill
The file
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.