Agent skill

Tilelang Example Merge

by tile-ai in tile-ai/tilelang-ascend

将算子的 kernel 文件和测试文件合并为单文件 example{op}.py,用于上库提交 PR. An agent skill from tile-ai/tilelang-ascend.

MITAuto-check passed

Install Tilelang Example Merge

skills CLI
$ npx skills add tile-ai/tilelang-ascend --skill tilelang-example-merge -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install tile-ai/tilelang-ascend tilelang-example-merge --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/tile-ai/tilelang-ascend.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.agents/skills/tilelang-example-merge .claude/skills/tilelang-example-merge && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
tilelang-example-merge
GitHub stars
402
Token cost
~2.6k tokens
SKILL.md length
632 words
Files
2
Skills in repo
20
Repo updated
First seen
Licence
MIT

At a glance

将算子的 kernel 文件和测试文件合并为单文件 example{op}.py,用于上库提交 PR. An agent skill from tile-ai/tilelang-ascend.

  • Works in 3 steps: 确认算子名(用户指定或从对话上下文提取) → 读取 examples/{op}/{op}.py,提取完整 kernel 代码 → 读取 examples/{op}/test_{op}.py,理解测试结构
  • SKILL.md covers 概述, 触发条件, 输入 and 工作流程, plus 4 more sections
  • Calls python

What it does

Tilelang Example Merge is an agent skill from tile-ai/tilelang-ascend. 将算子的 kernel 文件和测试文件合并为单文件 example{op}.py,用于上库提交 PR。 合并后的文件包含完整 kernel 实现 + 1 个代表性 L0 用例 + 1 个代表性 L1 用例, 全部内联,不依赖 import 兄弟模块。当用户提到上库、提交 PR、合并算子文件、 生成 example 文件、单文件提交、准备上库、repo submission、提 PR 前合并文件、 或需要把 kernel 和 test 合成一个文件时必须使用本 skill。即使用户没有明确说 "merge",只要意图是将算子代码整理成仓库可接收的单文件示例,也应触发。

Its SKILL.md is about 2.6k tokens, which your agent loads only when the skill is triggered. The skill folder holds 2 other files (for example `evals/evals.json`).

The repository describes itself as: Ascend TileLang adapter. The licence is MIT.

Example prompts

  • “/tilelang-example-merge”

Requirements

  • Python 3

Workflow steps

3 steps, taken from the first numbered list in SKILL.md.

  1. 确认算子名(用户指定或从对话上下文提取)
  2. 读取 examples/{op}/{op}.py,提取完整 kernel 代码
  3. 读取 examples/{op}/test_{op}.py,理解测试结构

What it can do on your machine

Read from SKILL.md and the folder at commit 83b0ece. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • python

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Tilelang Example Merge loads about 2.6k tokens when it runs. Until then it costs about 78 tokens; SKILL.md has 632 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~78
When it runs · the whole SKILL.md, loaded when a task matches
~2.6k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from tile-ai/tilelang-ascend at commit 83b0ece, republished under its MIT licence (© tile-ai). 632 words, ~2,611 tokens.

Download SKILL.mdSave it as .claude/skills/tilelang-example-merge/SKILL.md (or your agent's skills folder). This skill also uses 1 other file; get the full folder from GitHub.
name
tilelang-example-merge
description
将算子的 kernel 文件和测试文件合并为单文件 example_{op}.py,用于上库提交 PR。 合并后的文件包含完整 kernel 实现 + 1 个代表性 L0 用例 + 1 个代表性 L1 用例, 全部内联,不依赖 import 兄弟模块。当用户提到上库、提交 PR、合并算子文件、 生成 example 文件、单文件提交、准备上库、repo submission、提 PR 前合并文件、 或需要把 kernel 和 test 合成一个文件时必须使用本 skill。即使用户没有明确说 "merge",只要意图是将算子代码整理成仓库可接收的单文件示例,也应触发。

TileLang Example Merge

概述

将算子开发阶段的双文件结构({op}.py 纯 kernel + test_{op}.py 分层测试套件) 合并为仓库上库用的单文件 example_{op}.py。

为什么要合并:算子开发时 kernel 和测试分离便于迭代,但仓库上库只接收单文件 示例(参考 examples/normalization/layer_norm.py、examples/developer_mode/gelu_mul_developer.py 的惯例)。单文件示例自包含、可直接 python example_{op}.py 运行验证。

合并策略:kernel 完整保留 + 从测试套件中自动选取 1 个 L0 代表性用例 + 1 个 L1 代表性用例,精简辅助函数,使用 torch.testing.assert_close 做精度检查。

触发条件

  • 用户提到"上库"、"提交 PR"、"合并文件"、"生成 example"、"单文件提交"
  • 用户要把算子代码整理成仓库可接收的单文件示例
  • 用户提到 "example_softmax.py"、"example_layer_norm.py" 等命名模式

输入

参数说明示例
算子名算子目录名和文件名前缀softmax

输入文件(隐式从算子名推导):

  • examples/{op}/{op}.py — 纯 kernel 文件
  • examples/{op}/test_{op}.py — 分层测试文件

输出文件:

  • examples/{op}/example_{op}.py — 合并后的单文件示例

工作流程

第一步:读取源文件
  1. 确认算子名(用户指定或从对话上下文提取)
  2. 读取 examples/{op}/{op}.py,提取完整 kernel 代码
    • 包括模块级常量(pass_configs、CAST_MODE_* 等)
    • 包括 @tilelang.jit 装饰的函数及其内部的 @T.prim_func
    • 不要包含 if __name__ == "__main__" 块(如果有的话)
  3. 读取 examples/{op}/test_{op}.py,理解测试结构
    • 识别 L0 测试用例(通常在 test_{op}_l0() 函数或 test_configs 列表中)
    • 识别 L1 测试用例(通常在 test_{op}_l1() 函数或 L1_CASES 列表中)
    • 提取 golden 参考实现函数
    • 提取 get_precision 函数及 dtype→阈值映射表(用于第四步查表填占位符)
第二步:选取代表性用例
L0 代表性用例选取

L0 是门槛测试(规则 shape,block 整除),选取最具代表性的一个:

  1. 优先:名称含 "typical" 或 "standard" 的用例(如 l0_typical)
  2. 次选:shape 最大的用例(最大 N 或最大 B×N,最能代表真实工作负载)
  3. 兜底:第一个 L0 用例
L1 代表性用例选取

L1 是功能测试(含不规则 shape、数值范围覆盖),选取最标准的规则 shape 用例:

  1. 优先:带 D-SHAPE-ALIGNED tag 的用例(规则 shape,无尾块)
  2. 次选:第一个 shape 为规则对齐的用例(B % block_M == 0 且 N % block_N == 0)
  3. 兜底:第一个 L1 用例

选取时注意避开极端边界用例(如 B=1、N=1、超大数值范围),这些适合分层测试 但不适合作为上库示例的代表用例。

第三步:提取 golden 参考实现

从 test_{op}.py 中提取 golden 函数(通常名为 golden_{op} 或直接内联在测试中), 作为独立函数复制到 example_{op}.py 的 kernel 之后、if __name__ 块之前。

提取规则:

  1. 保留独立函数:不要内联到测试循环里。golden 函数放循环外,循环内调用 ref = golden(x)。这样 golden 逻辑只写一遍,多个用例复用,与 test_{op}.py 结构一致。
  2. 原样复制函数体:保留数学逻辑,去掉冗长 docstring(一行注释说明即可)。
  3. 不要重新实现:如果原 golden 调用了 PyTorch 内置函数(如 F.softmax、 torch.layer_norm),直接用该调用,不要手写等价实现,避免引入新 bug。
  4. 函数签名对齐:golden 函数的输入参数应与测试循环中传入的张量一致(通常是 def golden(x): return ...)。

例如 test_softmax.py 的 golden 是:

python
def golden_softmax(x):
    return torch.nn.functional.softmax(x.float(), dim=-1).to(x.dtype)

复制到 example_softmax.py 后保留为独立函数,循环内 ref = golden_softmax(x)。

精度阈值提取

精度阈值必须根据 test_configs 中选中用例的 dtype 动态确定,不能硬编码某个 dtype 的阈值。提取步骤:

  1. 确定选中 dtype:读取选中的 L0/L1 用例的 dtype 字段。
  2. 查 test_{op}.py 的 get_precision 表:找到该 dtype 对应的 (atol, rtol, max_abs_limit, required_ratio) 四元组。
  3. 填入模板占位符:将四个数值替换模板中的 {atol}、{rtol}、{max_abs_limit}、 {required_ratio}。

各 dtype 的标准阈值参考(源自 test_{op}.py 的 get_precision,与 tilelang-op-test-design/references/precision-standard.md 一致):

dtypeatolrtolmax_abs_limitrequired_ratio
float162**-142**-91e-10.99
bfloat162**-102**-61e00.99
float32 / "float"2**-162**-101e-20.99
hifloat322**-162**-101e-20.99
float8_e4m32**-42**-21e00.99
float8_e5m22**-32**-11e-10.99
int8/int16/int32/int64/uint80.00.00.01.0

注意:不同算子的 test_{op}.py 可能只覆盖表中部分 dtype。以目标算子 test 文件 中实际存在的为准,不要套用上表缺失的 dtype。

多 dtype 处理:如果选中用例存在多个不同 dtype(少见),不能在循环外写死一组 阈值。需在循环内按 dtype 分支查表,例如:

python
for B, N, block_M, block_N, dtype, level in test_configs:
    ...
    # Precision thresholds by dtype (from test_{op}.py get_precision)
    if dtype == "float16":
        atol, rtol, max_abs_limit, required_ratio = 2**-14, 2**-9, 1e-1, 0.99
    elif dtype == "float32" or dtype == "float":
        atol, rtol, max_abs_limit, required_ratio = 2**-16, 2**-10, 1e-2, 0.99
    # ... 只列选中用例实际涉及的 dtype
    ratio = (abs_err <= (atol + rtol * ref_cpu[m].abs())).float().mean().item()
    max_abs = abs_err.max().item()
    assert ratio >= required_ratio and max_abs <= max_abs_limit, ...

单 dtype(常见):如果选中用例 dtype 相同,直接在循环外写死该 dtype 的四个数值 字面量(不分支、不封装函数),保持代码精简。

第四步:生成合并文件

按以下模板生成 example_{op}.py(参考 examples/normalization/layer_norm.py 和 examples/developer_mode/gelu_mul_developer.py 的仓库惯例):

python
import tilelang
from tilelang import language as T
import torch

tilelang.cache.clear_cache()

# ========== Operator Implementation ==========
# (从 {op}.py 复制的 pass_configs、常量、@tilelang.jit 函数,原样保留)

pass_configs = {
    tilelang.PassConfigKey.TL_ASCEND_AUTO_SYNC: True,
    # ... 其他配置
}

@tilelang.jit(out_idx=[1], pass_configs=pass_configs)
def {op}(...):
    """{算子简述}"""
    # ... kernel 完整实现 ...
    return main


# ========== Golden reference ==========
def golden_{op}(x):
    """{一句话说明}"""
    return {test_{op}.py 中的 golden 函数体,原样复制}


# ========== Tests ==========
if __name__ == "__main__":
torch.manual_seed(0)

# Representative configs: 1 L0 + 1 L1
test_configs = [
    # (B, N, block_M, block_N, dtype, level)
    (..., ..., ..., ..., "...", "L0"),  # {L0 选中用例描述}
    (..., ..., ..., ..., "...", "L1"),  # {L1 选中用例描述}
]

for B, N, block_M, block_N, dtype, level in test_configs:
    print(f"Testing {op} {level} with B={B}, N={N}, block=({block_M},{block_N}), dtype={dtype}")
    func = {op}(B, N, block_M, block_N, dtype=dtype)
    print("Init successful!")
    torch_dtype = getattr(torch, dtype) if dtype != "float" else torch.float32
    x = torch.randn(B, N, dtype=torch_dtype).npu()
    y = func(x)
    ref = golden_{op}(x)
    # Precision check ({dtype} mixed tolerance, inlined — thresholds from test_{op}.py)
    y_cpu, ref_cpu = y.detach().cpu().float(), ref.detach().cpu().float()
    m = torch.isfinite(ref_cpu)
    abs_err = (y_cpu[m] - ref_cpu[m]).abs()
    ratio = (abs_err <= ({atol} + {rtol} * ref_cpu[m].abs())).float().mean().item()
    max_abs = abs_err.max().item()
    assert ratio >= {required_ratio} and max_abs <= {max_abs_limit}, f"precision fail: ratio={ratio:.4f} max_abs={max_abs:.3e}"
    print(f"Test pass! matched_ratio={ratio:.4f} max_abs={max_abs:.3e}")

print("Kernel Output Match!")
关键格式约定(必须遵循仓库惯例)
  1. 精度检查:内联混合容差检查,不要用 torch.testing.assert_close,也不要 封装成函数。直接在测试用例中按选中 dtype 的阈值内联计算。阈值根据 test_configs 中选中用例的 dtype 从 test_{op}.py 的 get_precision 表动态查取,填入模板的 {atol}/{rtol}/{max_abs_limit}/{required_ratio} 占位符——禁止硬编码某个 固定 dtype 的阈值。双门控:逐元素 |actual-golden| <= atol + rtol*|golden|, 整体 matched_ratio >= required_ratio 且 max_abs_error <= max_abs_limit。查表 规则和多 dtype 处理见上方"精度阈值提取"小节。
  2. 缓存清理:文件顶部用 tilelang.cache.clear_cache()(不是 tilelang.disable_cache()), 与仓库现有示例一致。
  3. 输入数据:用 torch.randn(...).npu() 生成随机输入。如果原测试用了特定数值范围 (如 uniform_(-1000, 1000)),L1 代表用例可以保留该范围,但 L0 用标准 randn。
  4. 打印格式:print(f"Testing {op} ... with ...") → print("Init successful!") → print("Test pass!") → 末尾 print("Kernel Output Match!"),与仓库现有示例一致。
  5. 无 import 兄弟模块:example_{op}.py 中禁止出现 from {op} import {op} 或 sys.path.insert 等导入语句。kernel 代码直接内联。
  6. 无分层测试框架:不要保留 --level 参数分发、COVERAGE_CATEGORY、 L1_CASES 列表、check_precision 等分层测试基础设施。上库示例用单个 test_configs 列表 + for 循环顺序执行两个代表性用例(1 个 L0 + 1 个 L1), 循环体内复用同一套 kernel 编译/运行/golden/精度检查逻辑,避免代码重复。 这与仓库现有示例(layer_norm.py、gelu_mul_developer.py 的 test_configs 循环)一致。
  7. 使用 if __name__ == "__main__" 守卫:测试代码放在 if __name__ == "__main__": 块中,python example_{op}.py 直接运行时会执行两个代表性用例并打印 PASS/FAIL。 这比仓库现有示例(layer_norm.py 等用模块级测试)更显式,且 import 时不自动执行。
Show full SKILL.md (198 more words)Show less
第五步:验证

生成文件后,运行验证:

bash
source set_env.sh
python examples/{op}/example_{op}.py

确认输出包含 "Kernel Output Match!"。如果失败,检查:

  • kernel 代码是否完整复制(漏了常量或辅助函数)
  • golden 实现是否正确
  • shape/dtype 是否与原测试一致
  • 是否有遗留的 import 语句

输出文件结构

生成的 example_{op}.py 分四段:

1. imports + tilelang.cache.clear_cache()
2. kernel 实现(pass_configs + @tilelang.jit 函数)   ← 从 {op}.py 复制
3. golden 参考实现(独立函数)                          ← 从 test_{op}.py 复制
4. if __name__ == "__main__": 测试代码(test_configs 循环 + 末尾打印)  ← 精简

目标行数:通常 80-150 行(kernel 行数 + 每个测试约 10-15 行)。

注意事项

  • 不要修改 kernel 逻辑:kernel 代码从 {op}.py 原样复制,不做任何改动。如果 kernel 依赖模块级辅助函数(如 cast_or_copy),一并复制。
  • 保留必要的模块级常量:pass_configs、CAST_MODE_*、VEC_NUM 等被 kernel 使用的常量必须保留。未被选中队列测试使用的常量(如 COVERAGE_MANIFEST)丢弃。
  • golden 简化但不失真:golden 函数保留正确的数学逻辑,但可以去掉冗长的 docstring。 如果原 golden 调用了 PyTorch 内置函数(如 F.softmax、torch.layer_norm),直接用 该调用,不要重新实现。
  • dtype 处理:如果 kernel 的 dtype 参数用 "float" 表示 float32,测试中需用 getattr(torch, dtype) if dtype != "float" else torch.float32 转换,与仓库惯例一致。
  • 原文件保留:生成 example_{op}.py 后,原 {op}.py 和 test_{op}.py 不删除, 它们仍用于开发阶段的分层测试。example_{op}.py 是上库用的精简单文件。

常见问题

测试文件结构不是标准 L0/L1 分层怎么办?

有些算子的测试文件可能用不同的结构(如单个 test_configs 列表无 L0/L1 区分)。 此时:

  • 将第一个规则 shape 用例作为 "L0 representative"
  • 将第二个规则 shape 用例(或稍大 shape 的用例)作为 "L1 representative"
  • 在注释中标注 "representative" 而非 "L0"/"L1"
kernel 文件有 if __name__ == "__main__" 块怎么办?

纯 kernel 文件({op}.py)通常没有 __main__ 块。如果有,只复制 kernel 部分 (imports + pass_configs + @tilelang.jit 函数),丢弃 __main__ 块。

算子有多个 kernel 函数怎么办?

如果 {op}.py 包含多个 @tilelang.jit 函数,全部保留(它们可能互相调用或用于 不同配置)。测试代码中调用主 kernel。

生成的文件跑不过怎么办?

最常见原因:

  1. 漏复制常量/辅助函数:检查 kernel 是否引用了未复制的模块级符号
  2. golden 与 kernel dtype 路径不一致:确保 golden 的 dtype 转换与 kernel 对齐
  3. shape 不匹配:确保测试 shape 与 kernel 的 jit 参数一致
  4. 缺少 torch.npu.synchronize():如果原测试有同步调用,保留它

参考示例

仓库中现有的单文件示例(合并后的目标格式参考):

  • examples/normalization/layer_norm.py — kernel + test_configs 循环 + assert_close
  • examples/developer_mode/gelu_mul_developer.py — kernel + test_configs 循环 + assert_close
  • examples/normalization/rms_norm.py — 同上模式

这些文件的共同特征:单文件、模块级测试、tilelang.cache.clear_cache()、末尾 print("Kernel Output Match!")。

与本 skill 的差异:仓库现有示例用 torch.testing.assert_close(rtol=1e-2, atol=1e-2) 做精度检查,但本 skill 按用户要求采用 test_{op}.py 的混合容差标准(内联,按 dtype 写死阈值),比 assert_close 的单一 rtol/atol 更贴合算子精度分级要求。其余格式 (单文件、clear_cache、末尾打印)保持与仓库惯例一致。

© tile-ai, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 1 other file in .agents/skills/tilelang-example-merge of tile-ai/tilelang-ascend.

  • SKILL.md
  • evals/evals.json

Open the folder on GitHubat commit 83b0ece

Compare with similar skills

Tilelang Example Merge next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Tilelang Example Merge compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Tilelang Example Merge this skilltile-ai/tilelang-ascend402—~2.6kAutomated safety check: PassMIT
Mergeremotion-dev/remotion63k—~508Automated safety check: PassCustom licence
Mergewithastro/astro63k—~153Automated safety check: PassCustom licence
Merge Upsymfony/symfony31k—~4kAutomated safety check: PassMIT
Mergealirezarezvani/claude-skills28k—~587Automated safety check: PassMIT
Kernel Organizationsgl-project/sglang37k—~1.3kAutomated safety check: PassApache-2.0

Similar skills

  • Merge

    remotion-dev/remotion

    Official

    Wait for a Remotion pull request to become mergeable, handle merge conflicts, distinguish genuine CI failures from flakes, rerun flaky checks through the flake skill, and merge the PR.

    63k GitHub stars~508 tokensUpdated today
    DevelopmentAuto-check passed
  • Merge

    withastro/astro

    Official

    Handle main-to-next merge tasks including conflict resolution, changeset cleanup, and CI fix-ups.

    63k GitHub stars~153 tokensUpdated today
    Auto-check passed
  • Merge Up

    symfony/symfony

    Cascade-merge maintained Symfony branches from oldest to newest (e.g.

    31k GitHub stars~4k tokensUpdated today
    DevelopmentAuto-check passed
  • Merge

    alirezarezvani/claude-skills

    Merge the winning agent's branch into base, archive losers, and clean up worktrees.

    28k GitHub stars~587 tokensUpdated 1 mo ago
    DevelopmentAuto-check passed
  • Kernel Organization

    sgl-project/sglang

    Apply the SGLang kernels RFC when adding, moving, splitting, or reviewing kernel APIs, registry metadata, kernel tests, benchmarks, and model-specific implementations.

    37k GitHub stars~1.3k tokensUpdated today
    AI & LLM EngineeringAuto-check passed
  • Merging PRs

    PostHog/posthog

    Official

    Merge a PR into master through the Trunk merge queue and babysit it until it lands.

    40k GitHub stars~4.1k tokensUpdated today
    Auto-check passed

More from tile-ai/tilelang-ascend

All 20 skills in this repo
  • Tilelang Submodule Pull

    tile-ai/tilelang-ascend

    Automatically pull tilelang repository and its third-party code.

    402 GitHub stars~1.5k tokensUpdated today
    Auto-check passed
  • Tilelang Debug Helper

    tile-ai/tilelang-ascend

    How to add debugging capabilities to TileLang Ascend example operators.

    402 GitHub stars~1.7k tokensUpdated today
    Auto-check: notes
  • Tilelang A5 Sim Convert

    tile-ai/tilelang-ascend

    将 tilelang example 脚本转换为可在 A5 camodel 仿真器上直接运行的版本。输入脚本路径,输出一个新的 sim.py 文件,不覆盖原始文件。触发:仿真运行、camodel、A5 仿真、sim 模式、转换脚本为仿真、不需要 NPU 跑 kernel、simulate A5。

    402 GitHub stars~1.1k tokensUpdated today
    Auto-check passed
  • Tilelang Env Check

    tile-ai/tilelang-ascend

    TileLang-Ascend 环境检查与配置验证技能。检查代码仓库完整性、编译安装状态、环境变量配置,并运行简单测试验证环境。发现问题会自动调用相关 skill 进行修复,并按依赖顺序重新执行后续步骤。触发关键词:"环境检查"、"检查环境"、"验证环境"、"环境配置"、"环境搭建"、"env check"、"check environment"、"verify…

    402 GitHub stars~3.4k tokensUpdated today
    Auto-check passed
  • Tilelang Op Test Design

    tile-ai/tilelang-ascend

    TileLang-Ascend 算子测试设计技能。支持多种场景:(1) 从 design.md 设计测试配置 (2) 从 examples/{op}/.py 补充测试 (3) 手动提供算子信息生成测试 (4) 测试覆盖率分析。理解算子实现逻辑后智能判断测试策略。触发:设计算子测试、生成测试用例、补充测试、测试覆盖率不足。

    402 GitHub stars~5.6k tokensUpdated today
    Auto-check passed
  • Tilelang Review

    tile-ai/tilelang-ascend

    检查代码格式是否符合 CI 规则。自动检测并安装缺失工具(ruff、clang-format),先运行检查生成报告,使用醒目方式询问用户后,仅在用户同意时执行修复。工作流程:检测环境→自动安装缺失工具→运行检查→生成报告→醒目询问→用户确认→执行修复。使用此技能当:用户要求"格式检查"、"格式化代码"、"代码格式化"、"检查代码格式"、"代码…

    402 GitHub stars~2.8k tokensUpdated today
    Auto-check: notes

Questions about Tilelang Example Merge

What does Tilelang Example Merge do?

将算子的 kernel 文件和测试文件合并为单文件 example{op}.py,用于上库提交 PR. An agent skill from tile-ai/tilelang-ascend. Tilelang Example Merge is an agent skill from tile-ai/tilelang-ascend.

How do I install Tilelang Example Merge in Claude Code?

Run `npx skills add tile-ai/tilelang-ascend --skill tilelang-example-merge -a claude-code`. Or copy the skill folder (.agents/skills/tilelang-example-merge in tile-ai/tilelang-ascend) into .claude/skills/tilelang-example-merge in your project. Claude Code loads it when a task matches its description.

How do I install Tilelang Example Merge in Codex?

Run `npx skills add tile-ai/tilelang-ascend --skill tilelang-example-merge -a codex`. Or copy the skill folder (.agents/skills/tilelang-example-merge in tile-ai/tilelang-ascend) into .agents/skills/tilelang-example-merge in your project. Codex loads it when a task matches its description.

Can I use Tilelang Example Merge in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add tile-ai/tilelang-ascend --skill tilelang-example-merge -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/tilelang-example-merge, .gemini/skills/tilelang-example-merge, .github/skills/tilelang-example-merge and .opencode/skills/tilelang-example-merge in your project.

What does Tilelang Example Merge need to run?

Going by SKILL.md and its folder, Tilelang Example Merge needs the command-line tools its instructions call (python). Our summary lists: Python 3.

Does Tilelang Example Merge access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Tilelang Example Merge safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Tilelang Example Merge use?

Tilelang Example Merge is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Tilelang Example Merge use?

About 2.6k tokens (SKILL.md is roughly 10k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Tilelang Example Merge?

Skills that share tags, products or a category with Tilelang Example Merge: Merge (remotion-dev/remotion, 63k stars), Merge (withastro/astro, 63k stars), Merge Up (symfony/symfony, 31k stars) and Merge (alirezarezvani/claude-skills, 28k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Tilelang Example Merge?

tile-ai (a GitHub organization) maintains it in tile-ai/tilelang-ascend, which has 402 GitHub stars. The repository holds 20 skills in this directory. The repository was last updated on October 9, 2026.

Source: tile-ai/tilelang-ascend on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.