---
name: kernel-triton-writing-5a379c6acccc
tags: [triton]
description: >
  ONLY for OpenAI Triton (@triton.jit) kernel development. NEVER use for
  CUDA C++ kernels, TileIR, or profiling tools (ncu, nsys).
  The user's request must involve Triton explicitly. Covers Triton-specific
  patterns: fused elementwise, reductions (softmax, LayerNorm, RMSNorm),
  tiled GEMM with triton.autotune, and flash attention. Workflow:
  design, write, verify (with fast-path for explicit requests).
license: Apache-2.0
metadata:
  author: NVIDIA Corporation
---

# Triton Kernel Writing

## Principles

### Correctness First

1. Never benchmark before verification passes.
2. Always mask loads and stores for non-divisible shapes.
3. Include `kernel_fn`, `reference_fn`, and `get_inputs()` exports for companion scripts.
4. Always run `scripts/verify_kernel.py` to validate against the reference.

### FP16/BF16 Precision Rules (LOW FREEDOM -- follow exactly)

Transcendental functions (`tl.exp`, `tl.log`, `tl.math.erf`, `tl.math.tanh`) require fp32 inputs.

```python
# WRONG -- compilation error or wrong results with fp16/bf16:
result = tl.exp(x)

# CORRECT -- cast to fp32, compute, cast back:
x_fp32 = x.to(tl.float32)
result = tl.exp(x_fp32).to(x.dtype)
```

Rule: any math function beyond basic arithmetic (+, -, *, /) requires fp32 cast in, original dtype cast out.

Additional precision constraints:
- `tl.sigmoid()` is unavailable in some Triton versions. Use `1.0 / (1.0 + tl.exp(-x_fp32))`.
- Always cast back to `x.dtype` before `tl.store` -- mismatches cause "Type mismatch, store Float32 to Float16".
- Unlike PyTorch, Triton does NOT auto-promote fp16/bf16 to fp32 for accumulation. Always use `tl.float32` accumulators for `tl.dot`.
- **TF32 for matmul:** On Ampere+/Hopper, `tl.dot` uses TF32 by default for fp32 inputs (same as `torch.mm`). Do NOT add `input_precision="ieee"` — it is 3-8x slower. TF32 is the correct default. If verification fails due to TF32 precision (~0.01-0.1 abs diff), ensure `reference_fn` also uses TF32 (plain `torch.mm`, no `allow_tf32=False`).

### CPU-GPU Sync Avoidance (LOW FREEDOM)

Never call `.item()` in kernel wrappers. It forces a CPU-GPU sync (~50-100us per call).

| Pitfall | Fix |
|---------|-----|
| `tensor.item()` for seed | `x.data_ptr() % (2**31)` |
| `torch.randint(...).item()` | Use tensor metadata for pseudo-random seed |
| Allocating output every call | Accept pre-allocated output as parameter |
| Python loops calling kernel | Batch operations |

### C Integer Division Semantics (CRITICAL)

Triton uses **C semantics** (round toward zero) for `//` and `%`, NOT Python semantics (round toward negative infinity). This only matters when operands can be negative.

| Expression | Python | Triton/C |
|------------|--------|----------|
| `-7 // 2` | `-4` | `-3` |
| `-7 % 2` | `1` | `-1` |

**Safe pattern:** Ensure all index/offset values are non-negative. If negative values are possible, use `(idx % BLOCK + BLOCK) % BLOCK`.

See [references/concepts-semantics.md](references/concepts-semantics.md) for full rules and scalar-only exception.

### Kernel Design Mental Model

- **Parallelization axis:** Element-wise kernels parallelize over flattened elements. Row-wise kernels (LayerNorm, softmax) parallelize over rows. Matmul kernels tile in 2D (M, N).
- **Block size:** Power-of-2 only (256, 512, 1024, 2048). Start with 1024 for H100, 512 for V100.
- **Memory coalescing:** Adjacent threads must access adjacent memory addresses. The compiler handles this automatically from block-level pointer arithmetic.
- **Grid:** Use `triton.cdiv(n_elements, BLOCK_SIZE)`. With autotune, grid must be a lambda: `lambda meta: (triton.cdiv(n, meta['BLOCK_SIZE']),)`.
- **Decorator order:** `@triton.autotune` (outermost) -> `@triton.heuristics` -> `@triton.jit` (innermost).
- **`reset_to_zero`:** Required for autotune on kernels that accumulate (e.g., matmul output). Without it, later configs see leftover values from earlier trials.

## Workflow

**Fast path:** If the user explicitly requests a Triton kernel (e.g., "Write a Triton kernel for X", "Implement softmax in Triton"), start at **Phase 2**. Only use Phase 0-1 when the request is ambiguous about whether Triton is appropriate.

### Phase 0: Route the Operator (only for ambiguous requests)

Skip this phase if the user explicitly asks for a Triton kernel. Only use when the request is ambiguous (e.g., "optimize this operation").

Triton wins when 2+ operations can share registers instead of writing/reading global memory. Quick rules:

| Pattern | Decision |
|---------|----------|
| Single element-wise op (`relu`, `sigmoid`) | SKIP — PyTorch already optimal |
| Standalone matmul | SKIP — cuBLAS is optimized |
| Standard attention | SKIP — Use FlashAttention |
| Element-wise chain (2+ ops), reduction, matmul + epilogue | USE TRITON |

If SKIP, recommend the alternative and STOP. See [references/operator-routing.md](references/operator-routing.md) for edge cases.

### Phase 1: Analyze the Operator (only for ambiguous requests)

From the user's request, identify: (1) operation type, (2) parallelization strategy, (3) input shapes and dtypes.

### Phase 2: Design the Kernel

Pick the skeleton below that matches your operation. **These skeletons are sufficient for element-wise, reduction, matmul, and fusion kernels — do NOT read reference files for these common patterns.** Only consult `references/` when implementing uncommon patterns (grouped GEMM, TMA, extern functions) or debugging issues.

**Element-wise skeleton** (GELU, dropout, fused ops on flat tensors):

```python
@triton.jit
def kernel(x_ptr, out_ptr, n_elements, BLOCK_SIZE: tl.constexpr):
    pid = tl.program_id(0)
    offsets = pid * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE)
    mask = offsets < n_elements
    x = tl.load(x_ptr + offsets, mask=mask)
    # ... compute ...
    tl.store(out_ptr + offsets, result, mask=mask)
```

**Row-wise skeleton** (softmax, LayerNorm, RMSNorm — one program per row):

```python
@triton.jit
def kernel(x_ptr, out_ptr, n_cols, BLOCK_SIZE: tl.constexpr):
    row_idx = tl.program_id(0)
    col_offsets = tl.arange(0, BLOCK_SIZE)
    mask = col_offsets < n_cols
    x = tl.load(x_ptr + row_idx * n_cols + col_offsets, mask=mask, other=0.0)
    # ... reduce / normalize ...
    tl.store(out_ptr + row_idx * n_cols + col_offsets, result, mask=mask)
```

**Tiled matmul skeleton** (GEMM with 2D tiling, grouped ordering, and autotune):

```python
@triton.autotune(
    configs=[
        triton.Config({'BLOCK_M': 128, 'BLOCK_N': 256, 'BLOCK_K': 64, 'GROUP_SIZE_M': 8}, num_warps=8, num_stages=3),
        triton.Config({'BLOCK_M': 64, 'BLOCK_N': 256, 'BLOCK_K': 32, 'GROUP_SIZE_M': 8}, num_warps=4, num_stages=4),
        triton.Config({'BLOCK_M': 128, 'BLOCK_N': 128, 'BLOCK_K': 32, 'GROUP_SIZE_M': 8}, num_warps=4, num_stages=4),
        triton.Config({'BLOCK_M': 256, 'BLOCK_N': 64, 'BLOCK_K': 32, 'GROUP_SIZE_M': 8}, num_warps=4, num_stages=4),
    ],
    key=['M', 'N', 'K'],
)
@triton.jit
def matmul_kernel(
    a_ptr, b_ptr, c_ptr, M, N, K,
    stride_am, stride_ak, stride_bk, stride_bn, stride_cm, stride_cn,
    BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr,
    GROUP_SIZE_M: tl.constexpr,
):
    pid = tl.program_id(0)
    num_m_blocks = tl.cdiv(M, BLOCK_M)
    num_n_blocks = tl.cdiv(N, BLOCK_N)
    # Grouped ordering for L2 cache locality
    num_pid_in_group = GROUP_SIZE_M * num_n_blocks
    group_id = pid // num_pid_in_group
    first_pid_m = group_id * GROUP_SIZE_M
    group_size_m = min(num_m_blocks - first_pid_m, GROUP_SIZE_M)
    pid_m = first_pid_m + ((pid % num_pid_in_group) % group_size_m)
    pid_n = (pid % num_pid_in_group) // group_size_m

    offs_m = pid_m * BLOCK_M + tl.arange(0, BLOCK_M)
    offs_n = pid_n * BLOCK_N + tl.arange(0, BLOCK_N)
    offs_k = tl.arange(0, BLOCK_K)

    a_ptrs = a_ptr + offs_m[:, None] * stride_am + offs_k[None, :] * stride_ak
    b_ptrs = b_ptr + offs_k[:, None] * stride_bk + offs_n[None, :] * stride_bn
    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=tl.float32)

    for k in range(0, tl.cdiv(K, BLOCK_K)):
        a_mask = (offs_m[:, None] < M) & (offs_k[None, :] < K)
        b_mask = (offs_k[:, None] < K) & (offs_n[None, :] < N)
        a = tl.load(a_ptrs, mask=a_mask, other=0.0)
        b = tl.load(b_ptrs, mask=b_mask, other=0.0)
        acc += tl.dot(a, b)
        a_ptrs += BLOCK_K * stride_ak
        b_ptrs += BLOCK_K * stride_bk
        offs_k += BLOCK_K

    c_mask = (offs_m[:, None] < M) & (offs_n[None, :] < N)
    c_ptrs = c_ptr + offs_m[:, None] * stride_cm + offs_n[None, :] * stride_cn
    tl.store(c_ptrs, acc.to(c_ptr.dtype.element_ty), mask=c_mask)
```

### Phase 3: Write the Kernel

Create an output directory, then write the kernel file to `{output_dir}/kernel.py`.

The kernel file MUST include:
- `@triton.jit` decorated kernel function
- `@triton.autotune` for production kernels (see [references/api-core.md](references/api-core.md))
- Python wrapper function (descriptive name for external import)
- **Fixed contract exports** (companion scripts rely on these exact names):
  - `kernel_fn` — alias to the wrapper function
  - `reference_fn(*args)` — PyTorch reference with identical signature
  - `get_inputs()` — returns `list` of fresh CUDA tensors for testing/benchmarking

Concise example (fused GELU + dropout):

```python
import triton
import triton.language as tl
import torch

@triton.autotune(
    configs=[
        triton.Config({'BLOCK_SIZE': 1024}, num_warps=4),
        triton.Config({'BLOCK_SIZE': 2048}, num_warps=8),
    ],
    key=['n_elements'],
)
@triton.jit
def fused_gelu_dropout_kernel(
    x_ptr, out_ptr, n_elements, p, seed,
    BLOCK_SIZE: tl.constexpr,
):
    pid = tl.program_id(0)
    offsets = pid * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE)
    mask = offsets < n_elements

    x = tl.load(x_ptr + offsets, mask=mask)
    x_fp32 = x.to(tl.float32)
    x = (0.5 * x_fp32 * (1.0 + tl.math.erf(x_fp32 * 0.7071067811865476))).to(x.dtype)

    random = tl.rand(seed, offsets)
    x = tl.where(random > p, x / (1.0 - p), 0.0)

    tl.store(out_ptr + offsets, x, mask=mask)


def fused_gelu_dropout_triton(x: torch.Tensor, p: float = 0.1) -> torch.Tensor:
    n_elements = x.numel()
    out = torch.empty_like(x)
    grid = lambda meta: (triton.cdiv(n_elements, meta['BLOCK_SIZE']),)
    seed = (x.data_ptr() % (2**31)) ^ n_elements  # sync-free seed
    fused_gelu_dropout_kernel[grid](x, out, n_elements, p, seed)
    return out


# --- Fixed contract (companion scripts rely on these names) ---
kernel_fn = fused_gelu_dropout_triton

def reference_fn(x, p=0.1):
    torch.manual_seed((x.data_ptr() % (2**31)) ^ x.numel())
    return torch.nn.functional.dropout(
        torch.nn.functional.gelu(x), p, training=True
    )

def get_inputs():
    return [torch.randn(128 * 1024 * 1024, device="cuda")]
```

For more patterns (SiLU+mul, RMSNorm, linear+GELU, add+LayerNorm), see [references/patterns-fusion.md](references/patterns-fusion.md). For GEMM patterns, see [references/patterns-gemm.md](references/patterns-gemm.md).

### Phase 4: Verify Correctness

Run the companion verification script:

```bash
python scripts/verify_kernel.py {output_dir}/kernel.py --rtol 1e-3 --atol 1e-3
```

Output:
```json
{"correct": true, "max_abs_diff": 1.2e-7, "max_rel_diff": 3.4e-6, "details": "..."}
```

**Stop if `correct: false`.** Fix the kernel before benchmarking.

**Tolerance guide:**

| Dtype | rtol | atol | Notes |
|-------|------|------|-------|
| float16 | 1e-3 | 1e-3 | |
| bfloat16 | 1e-2 | 1e-2 | |
| float32 | 1e-5 | 1e-5 | Element-wise ops |
| float32 (matmul) | 1e-2 | 1e-1 | TF32 accumulation order differs between Triton tiles and cuBLAS |

### Phase 5: Benchmark Performance (optional)

Only benchmark if the user explicitly requests performance numbers. Skip this phase for correctness-focused requests.

```bash
python scripts/benchmark_kernel.py {output_dir}/kernel.py
```

Output:
```json
{"kernel_time_ms": 0.45, "reference_time_ms": 1.23, "speedup": 2.73, "warmup_iters": 10, "benchmark_iters": 40}
```

## References (consult only when stuck)

The skeletons and principles above cover element-wise, reduction, matmul, and fusion kernels. **Do NOT read reference files for these common patterns.**

Only consult `references/` when:
- Implementing **uncommon patterns** (grouped GEMM, TMA, persistent matmul, extern functions)
- **Debugging** a compile error or incorrect result not covered by the error table below
- Needing **API details** for an unfamiliar `tl.*` operation

**How to search:** Grep for your keyword across `references/`. Read only the file Grep points to.

| File | When to use |
|---|---|
| `references/api-core.md` | Unfamiliar `triton.autotune` / `triton.Config` options |
| `references/api-language.md` | Unfamiliar `tl.*` operations |
| `references/patterns-gemm.md` | Grouped GEMM, persistent matmul, TMA, MX formats |
| `references/patterns-advanced.md` | Flash attention details, backward passes, libdevice |
| `references/troubleshooting.md` | Debug ops, interpreter mode, env vars |

## Error Handling and Troubleshooting

### Common Errors

| Error / Symptom | Cause | Fix |
|---------|-------|-----|
| "Type mismatch, store Float32 to Float16" | Missing `.to(x.dtype)` before store | Cast fp32 result back |
| `BLOCK_SIZE is not a constexpr` | Block size passed as runtime value | Add `: tl.constexpr` annotation |
| `shape mismatch` in binary op | Tensor shapes don't broadcast | Check with `tl.static_print`; use `[:, None]` / `[None, :]` |
| Large diffs everywhere | Wrong dtype in `tl.load` | Check load dtype matches input |
| Matmul 3-8x slower than expected | `input_precision="ieee"` on `tl.dot` | Remove it; use TF32 default. Ensure `reference_fn` also uses TF32 |
| Matmul ~0.01-0.1 abs diff vs reference | TF32 vs IEEE mismatch | Use same precision in both kernel and reference (TF32 for both) |
| Diffs at boundaries | Missing mask | Add mask to all load/store ops |
| Random diffs | Race condition | Check atomics and ordering |
| NaN/Inf | Division by zero or fp16 overflow | Guard with epsilon; use `tl.float32` accumulator |
| `grid must be a tuple` | Grid lambda returns int, not tuple | Return `(value,)` with trailing comma |
| `expected constexpr` in `tl.arange` | Non-constexpr argument | Both args of `tl.arange(start, end)` must be constexpr |
| `triton.OutOfResources` | Register/shared memory pressure | Reduce BLOCK_SIZE or `num_stages` |
| Kernel not updating after edit | Stale compilation cache | `rm -rf ~/.triton/cache/` |
| Mismatched results vs PyTorch | C integer division semantics | Triton uses truncation; see `references/concepts-semantics.md` |

For extended error table, interpreter mode issues, and environment variables, see [references/troubleshooting.md](references/troubleshooting.md).

### When to Abort

Stop and report failure if:
1. **Not a good fit** -- Pure matmul or complex control flow (Phase 0 should catch this).
2. **Verification fails after 3 attempts** -- Numerical issues too severe to fix.
3. **No speedup** -- Reference is already well-optimized (cuBLAS, cuDNN).
4. **Hardware mismatch** -- Target GPU not available for testing.


---
Source attribution: yo-steven
Original project: https://github.com/yo-steven/skills-exploration-20260522/blob/6e1127ffb7690aa8038b172d0ae7cc720754870d/skills/TensorRT-LLM/kernel-triton-writing/SKILL.md
License: Apache-2.0
License information: https://www.apache.org/licenses/LICENSE-2.0
Original upstream instructions; downloaded from Mega AI Hub.

## Upstream license notice

SPDX-License-Identifier: Apache-2.0 AND CC-BY-4.0

Copyright (c) 2026 NVIDIA Corporation

This project is licensed under a dual license:
- Apache License 2.0 (for code and configuration)
- Creative Commons Attribution 4.0 International (CC BY 4.0) (for documentation)

You may use the contents of this repository under the terms of either license,
or both, as applicable to the type of content.

================================================================================
                                 Apache License
                           Version 2.0, January 2004
                        http://www.apache.org/licenses/

   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION

   1. Definitions.

      "License" shall mean the terms and conditions for use, reproduction,
      and distribution as defined by Sections 1 through 9 of this document.

      "Licensor" shall mean the copyright owner or entity authorized by
      the copyright owner that is granting the License.

      "Legal Entity" shall mean the union of the acting entity and all
      other entities that control, are controlled by, or are under common
      control with that entity. For the purposes of this definition,
      "control" means (i) the power, direct or indirect, to cause the
      direction or management of such entity, whether by contract or
      otherwise, or (ii) ownership of fifty percent (50%) or more of the
      outstanding shares, or (iii) beneficial ownership of such entity.

      "You" (or "Your") shall mean an individual or Legal Entity
      exercising permissions granted by this License.

      "Source" form shall mean the preferred form for making modifications,
      including but not limited to software source code, documentation
      source, and configuration files.

      "Object" form shall mean any form resulting from mechanical
      transformation or translation of a Source form, including but
      not limited to compiled object code, generated documentation,
      and conversions to other media types.

      "Work" shall mean the work of authorship, whether in Source or
      Object form, made available under the License, as indicated by a
      copyright notice that is included in or attached to the work
      (an example is provided in the Appendix below).

      "Derivative Works" shall mean any work, whether in Source or Object
      form, that is based on (or derived from) the Work and for which the
      editorial revisions, annotations, elaborations, or other modifications
      represent, as a whole, an original work of authorship. For the purposes
      of this License, Derivative Works shall not include works that remain
      separable from, or merely link (or bind by name) to the interfaces of,
      the Work and Derivative Works thereof.

      "Contribution" shall mean any work of authorship, including
      the original version of the Work and any modifications or additions
      to that Work or Derivative Works thereof, that is intentionally
      submitted to the Licensor for inclusion in the Work by the copyright owner
      or by an individual or Legal Entity authorized to submit on behalf of
      the copyright owner. For the purposes of this definition, "submitted"
      means any form of electronic, verbal, or written communication sent
      to the Licensor or its representatives, including but not limited to
      communication on electronic mailing lists, source code control systems,
      and issue tracking systems that are managed by, or on behalf of, the
      Licensor for the purpose of discussing and improving the Work, but
      excluding communication that is conspicuously marked or otherwise
      designated in writing by the copyright owner as "Not a Contribution."

      "Contributor" shall mean Licensor and any individual or Legal Entity
      on behalf of whom a Contribution has been received by the Licensor and
      subsequently incorporated within the Work.

   2. Grant of Copyright License. Subject to the terms and conditions of
      this License, each Contributor hereby grants to You a perpetual,
      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
      copyright license to reproduce, prepare Derivative Works of,
      publicly display, publicly perform, sublicense, and distribute the
      Work and such Derivative Works in Source or Object form.

   3. Grant of Patent License. Subject to the terms and conditions of
      this License, each Contributor hereby grants to You a perpetual,
      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
      (except as stated in this section) patent license to make, have made,
      use, offer to sell, sell, import, and otherwise transfer the Work,
      where such license applies only to those patent claims licensable
      by such Contributor that are necessarily infringed by their
      Contribution(s) alone or by combination of their Contribution(s)
      with the Work to which such Contribution(s) was submitted. If You
      institute patent litigation against any entity (including a
      cross-claim or counterclaim in a lawsuit) alleging that the Work
      or a Contribution incorporated within the Work constitutes direct
      or contributory patent infringement, then any patent licenses
      granted to You under this License for that Work shall terminate
      as of the date such litigation is filed.

   4. Redistribution. You may reproduce and distribute copies of the
      Work or Derivative Works thereof in any medium, with or without
      modifications, and in Source or Object form, provided that You
      meet the following conditions:

      (a) You must give any other recipients of the Work or
          Derivative Works a copy of this License; and

      (b) You must cause any modified files to carry prominent notices
          stating that You changed the files; and

      (c) You must retain, in the Source form of any Derivative Works
          that You distribute, all copyright, patent, trademark, and
          attribution notices from the Source form of the Work,
          excluding those notices that do not pertain to any part of
          the Derivative Works; and

      (d) If the Work includes a "NOTICE" text file as part of its
          distribution, then any Derivative Works that You distribute must
          include a readable copy of the attribution notices contained
          within such NOTICE file, excluding any notices that do not
          pertain to any part of the Derivative Works, in at least one
          of the following places: within a NOTICE text file distributed
          as part of the Derivative Works; within the Source form or
          documentation, if provided along with the Derivative Works; or,
          within a display generated by the Derivative Works, if and
          wherever such third-party notices normally appear. The contents
          of the NOTICE file are for informational purposes only and
          do not modify the License. You may add Your own attribution
          notices within Derivative Works that You distribute, alongside
          or as an addendum to the NOTICE text from the Work, provided
          that such additional attribution notices cannot be construed
          as modifying the License.

      You may add Your own copyright statement to Your modifications and
      may provide additional or different license terms and conditions
      for use, reproduction, or distribution of Your modifications, or
      for any such Derivative Works as a whole, provided Your use,
      reproduction, and distribution of the Work otherwise complies with
      the conditions stated in this License.

   5. Submission of Contributions. Unless You explicitly state otherwise,
      any Contribution intentionally submitted for inclusion in the Work
      by You to the Licensor shall be under the terms and conditions of
      this License, without any additional terms or conditions.
      Notwithstanding the above, nothing herein shall supersede or modify
      the terms of any separate license agreement you may have executed
      with Licensor regarding such Contributions.

   6. Trademarks. This License does not grant permission to use the trade
      names, trademarks, service marks, or product names of the Licensor,
      except as required for reasonable and customary use in describing the
      origin of the Work and reproducing the content of the NOTICE file.

   7. Disclaimer of Warranty. Unless required by applicable law or
      agreed to in writing, Licensor provides the Work (and each
      Contributor provides its Contributions) on an "AS IS" BASIS,
      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
      implied, including, without limitation, any warranties or conditions
      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
      PARTICULAR PURPOSE. You are solely responsible for determining the
      appropriateness of using or redistributing the Work and assume any
      risks associated with Your exercise of permissions under this License.

   8. Limitation of Liability. In no event and under no legal theory,
      whether in tort (including negligence), contract, or otherwise,
      unless required by applicable law (such as deliberate and grossly
      negligent acts) or agreed to in writing, shall any Contributor be
      liable to You for damages, including any direct, indirect, special,
      incidental, or consequential damages of any character arising as a
      result of this License or out of the use or inability to use the
      Work (including but not limited to damages for loss of goodwill,
      work stoppage, computer failure or malfunction, or any and all
      other commercial damages or losses), even if such Contributor
      has been advised of the possibility of such damages.

   9. Accepting Warranty or Additional Liability. While redistributing
      the Work or Derivative Works thereof, You may choose to offer,
      and charge a fee for, acceptance of support, warranty, indemnity,
      or other liability obligations and/or rights consistent with this
      License. However, in accepting such obligations, You may act only
      on Your own behalf and on Your sole responsibility, not on behalf
      of any other Contributor, and only if You agree to indemnify,
      defend, and hold each Contributor harmless for any liability
      incurred by, or claims asserted against, such Contributor by reason
      of your accepting any such warranty or additional liability.

   END OF TERMS AND CONDITIONS

================================================================================
              Creative Commons Attribution 4.0 International License
                              (CC BY 4.0)

                https://creativecommons.org/licenses/by/4.0/

You are free to:

  Share — copy and redistribute the material in any medium or format for any
  purpose, even commercially.

  Adapt — remix, transform, and build upon the material for any purpose, even
  commercially.

  The licensor cannot revoke these freedoms as long as you follow the license
  terms.

Under the following terms:

  Attribution — You must give appropriate credit, provide a link to the license,
  and indicate if changes were made. You may do so in any reasonable manner, but
  not in any way that suggests the licensor endorses you or your use.

  No additional restrictions — You may not apply legal terms or technological
  measures that legally restrict others from doing anything the license permits.

Notices:

  You do not have to comply with the license for elements of the material in the
  public domain or where your use is permitted by an applicable exception or
  limitation.

  No warranties are given. The license may not give you all of the permissions
  necessary for your intended use. For example, other rights such as publicity,
  privacy, or moral rights may limit how you use the material.

For the full legal text, see:
https://creativecommons.org/licenses/by/4.0/legalcode



# Mega AI Hub resource metadata

Creator / publisher: yo-steven
Original project: https://github.com/yo-steven/skills-exploration-20260522/blob/6e1127ffb7690aa8038b172d0ae7cc720754870d/skills/TensorRT-LLM/kernel-triton-writing/SKILL.md
Attribution basis: Source repository record

Fields of work: Data & Analytics, Development & Engineering
Roles: Backend Developer, Data Scientist, ML / AI Engineer
Specialties: Backend
Tools: Triton
Country applicability: Country not specified
Classification basis: Source catalog labels
Instruction SHA-256: 65912058c2e0ed1cf42bb9cf25ca7765d6e119c686815f4004c1646da864938f