cuTile GPU kernel programming: tile-based kernels with validation and orchestration workflows
Step-by-step kernel design (tile sizes, type annotations, launch), mandatory compile-run-validate loop, example-first workflow, and a deep agent-orchestration pipeline for multi-kernel tasks like transformer blocks
- What
- Step-by-step kernel design (tile sizes, type annotations, launch), mandatory compile-run-validate loop, example-first workflow, and a deep agent-orchestration pipeline for multi-kernel tasks like transformer blocks
- Cost
- Free
- Needs
- an NVIDIA GPU with cuTile installed for validating kernels — the skill generates and runs real GPU code
- Install
- Copy the installer prompt below into your Muse — your agent does the rest.
Curated by Skill Harbor — @nvidia's cuTile programming skill: the full workflow for writing high-performance GPU kernels in cuTile's tile-based Python DSL — search TileGym's `src/tilegym/ops/cutile/` for existing examples BEFORE implementing (then the packaged `examples/`), assess complexity (single kernel → simple workflow; 3+ kernels or inter-kernel dependencies → deep agent-orchestration pipeline with Op Tracer → Analyzer → parallel Kernel Agents → Composer), follow the mandatory steps (understand the problem, design tile/block sizes as powers of 2, annotate every constant with `ct.Constant[type]`, keep the forward path pure cuTile with `@ct.kernel` + `ct.launch` — no `nn.*`/`F.*` compute in `forward()`), and run the mandatory validation loop (generate → execute → fix, up to 3 attempts) against a PyTorch reference. Four critical rules anchor correctness: pure cuTile forward path, tile indices not element indices, power-of-2 tile dimensions, typed constants. Honest caveats: needs an NVIDIA GPU with cuTile installed to validate anything; a specialized skill for kernel authors, not for general Python users; the SKILL.md frontmatter also names CC-BY-4.0 alongside Apache-2.0 (license field varies by source — review before redistributing). Skill Harbor never reviews the code, review it yourself before use. Discovered via skills.sh.
Version:
Quick answer
cuTile GPU kernel programming is a free Muse skill for tile-based kernel design, step by step. It covers tile sizes, type annotations and launch, with a mandatory compile-run-validate loop and an example-first workflow. It also adds a deep agent-orchestration pipeline for multi-kernel tasks such as transformer blocks.
Install
Copy the install package below, then paste it into Muse🧪 Install ran in our test sandbox · Nov 18, 2026📌 Pinned version: commit 6d35235Step 0 — before installing anything: read this install prompt, summarize in a few lines what this build does and what installing it involves, then ask me to confirm before you proceed. Prerequisites: an NVIDIA GPU with cuTile installed for validating kernels — the skill generates and runs real GPU code Install "cuTile GPU kernel programming: tile-based kernels with validation and orchestration workflows" for me. It gives my agent @nvidia's cuTile workflow: search TileGym examples first, assess complexity to pick the simple or deep-orchestration workflow, design kernels step by step (power-of-2 tile sizes, typed constants, pure cuTile forward path), and run the mandatory generate→execute→validate loop against a PyTorch reference. Apache-2.0 licensed (note: the SKILL.md frontmatter also names CC-BY-4.0 — review the license field before redistributing). Repository: https://raw.githubusercontent.com/nvidia/skills/6d3523563914a577e26425854988c7d3f02ed0b7/skills/tilegym-cutile-python/SKILL.md Pinned version: install the code at commit 6d3523563914a577e26425854988c7d3f02ed0b7 exactly (Skill Harbor tested this version). Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Fetch the SKILL.md file (and any helper files) from the repository path into a temporary folder 2. Safety check: review the SKILL.md and scripts for anything suspicious (unexpected network calls, shell commands, credential harvesting). This repo should contain zero secrets in code, credentials only via the secure vault, allowed hosts declared in the SKILL.md. Verify that holds here; STOP on any red flag and tell me. 3. Install it as a skill: copy SKILL.md and its helper files into the agent's skills directory, in a folder named "tilegym-cutile-python". 4. Verify with no network calls: frontmatter valid, files in place. 5. Report what was installed, where, and what I still need to do myself (e.g. confirm cuTile is installed on an NVIDIA GPU accessible to the agent and tell it what kernel to write — shapes, dtypes, performance targets; clarify the approach for ambiguous optimization requests before coding). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.
Saved to your recent installs. Find it anytime on /connect.
🔒 Security, provenance and what this touches
The install prompt below already includes the vetting steps: your agent follows the community checklist before installing anything with executable code. Want more?
Verification certificate
Dated facts about what was checked for this listing, and how. This is not a safety certification.
- Pinned version
- 6d35235 · 2026-10-13
- Sandbox install test
- Install ran in our test sandbox · Nov 18, 2026
- Creator identity
- Not verified
- Fingerprint watch
- Last checked · Nov 19, 2026 : no change detected in the files linked by this install prompt
- Install prompt analysis
- 2026-10-13 : No commands, network calls, packages, file writes or API keys detected.
Skill Harbor never reviews the code of a listed skill. These are dated facts about the install prompt and our install test, not a safety certification. Review a skill yourself before you install it. How we verify
Questions
How do I install a skill?
Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the skill up for you — no manual configuration.
Where does my money go?
Straight to the seller. Skill Harbor never processes payments: checkout happens on the seller’s own page, usually Stripe.
What does the ✓ next to a creator’s name mean?
It means we confirmed the identity of the person behind the listing. It says nothing about the code itself — always check a skill before installing it.