triton-ascend-case-reduction-weighted-swiglu

triton-ascend-case-reduction-weighted-swiglu is a skill for Claude Code, Codex from mindspore-ai/akg. It costs 97 tokens per session (780 once invoked), scanned A, original, Apache-2.0.

A Triton Ascend guide for optimizing the backward pass of Weighted SwiGLU, a neural-network operation that applies weights and computes gradients.

In plain words
What is it for?
Use it when tuning a fused elementwise-and-reduction kernel for large 3D neural-network tensors on Ascend hardware.
Why use it?
It shows how reshaping a 3D tensor and splitting rows can simplify parallel work and improve memory access during a reduction.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one.

Good fit Use it when tuning a fused elementwise-and-reduction kernel for large 3D neural-network tensors on Ascend hardware.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/mindspore-ai/akg/triton-ascend-case-reduction-weighted-swiglu
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add mindspore-ai/akg --skill triton-ascend-case-reduction-weighted-swiglu
Clone the repo
git clone --depth 1 https://github.com/mindspore-ai/akg

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for triton-ascend-case-reduction-weighted-swiglu

README.md
[![agentmods](https://agentmods.dev/badge/skills/mindspore-ai/akg/triton-ascend-case-reduction-weighted-swiglu/github.svg)](https://agentmods.dev/skills/mindspore-ai/akg/triton-ascend-case-reduction-weighted-swiglu)
Your own site
<a href="https://agentmods.dev/skills/mindspore-ai/akg/triton-ascend-case-reduction-weighted-swiglu"><img src="https://agentmods.dev/badge/skills/mindspore-ai/akg/triton-ascend-case-reduction-weighted-swiglu/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for triton-ascend-case-reduction-weighted-swiglu

Your own site · 80×15
<a href="https://agentmods.dev/skills/mindspore-ai/akg/triton-ascend-case-reduction-weighted-swiglu"><img src="https://agentmods.dev/badge/skills/mindspore-ai/akg/triton-ascend-case-reduction-weighted-swiglu.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 97 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 780 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector pass 7 Sept 2026
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00097 $0.00780
Opus 5 $0.00048 $0.00390
Sonnet 5 $0.00019 $0.00156
Haiku 4.5 $0.00010 $0.00078

Measured 9d ago against content hash c6e22c848913, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-12, from the pricing page.

Security

Grade A, and why

triton-ascend-case-reduction-weighted-swiglu scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

akg_agents/python/akg_agents/op/resources/skills/triton-ascend/cases/triton-ascend-case-reduction-weighted-swiglu/SKILL.md · 65 lines

What it actually says

Weighted SwiGLU Backward 融合算子优化

任务特征

  • 数据尺寸:(16, 1024, 2048) × 3,3D融合算子
  • 特点:先逐元素操作,再reduce最后一根轴

优化 1:Reshape 降维

# 将前两个维度(B, M)合并为单个维度BM
x_reshaped = x.reshape(BM, N)
weight_reshaped = weight.reshape(BM, N)
grad_reshaped = grad.reshape(BM, N)

weighted_x_reshaped = weighted_x.reshape(BM, N)
grad_weight_reshaped = grad_weight.reshape(BM, N)
grad_x_reshaped = grad_x.reshape(BM)

优势:简化并行策略,优化内存访问模式,提高内核执行效率。

优化 2:行二次切分

for bm_start in range(0, BLOCK_SIZE_BM, SUB_BLOCK_SIZE_BM):
    bm_offsets = pid * BLOCK_SIZE_BM + bm_start + tl.arange(0, SUB_BLOCK_SIZE_BM)
    bm_mask = bm_offsets < BM

Autotune 配置

# (AI core=40)
# 1. grid=512>40,reduce轴切分较小,UB占满 -> 1105.84 us
triton.Config({'BLOCK_SIZE_BM': 32, 'SUB_BLOCK_SIZE_BM': 32, 'BLOCK_SIZE_N': 128})

# 2. grid=1024>40,reduce轴切分增至256 -> 1110.47 us
triton.Config({'BLOCK_SIZE_BM': 16, 'SUB_BLOCK_SIZE_BM': 16, 'BLOCK_SIZE_N': 256})

# 3. grid=2048>40,reduce轴切分增至512 -> 1091.26 us 最优
triton.Config({'BLOCK_SIZE_BM': 8, 'SUB_BLOCK_SIZE_BM': 8, 'BLOCK_SIZE_N': 512})

# 4. grid=32<40,reduce轴切分较大,UB占满 -> 1098.53 us
triton.Config({'BLOCK_SIZE_BM': 512, 'SUB_BLOCK_SIZE_BM': 8, 'BLOCK_SIZE_N': 512})

# 5. grid=40,有尾块,reduce轴切分较大,UB占满 -> 1094.60 us
triton.Config({'BLOCK_SIZE_BM': 416, 'SUB_BLOCK_SIZE_BM': 8, 'BLOCK_SIZE_N': 512})

总结

  1. Reshape降维可简化并行策略,优化内存访问
  2. 在优先占满UB前提下为reduce轴分配较大切分尺寸
  3. Grid数较大时,可能性能更优(配置3)
Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 9d ago First seen · 65 lines · 97 tokens per session scan A c6e22c848913

Subscribe to this mod's changes

triton-ascend-case-reduction-weighted-swiglu is a skill published in the GitHub repository mindspore-ai/akg (259 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 97 tokens to every session and 780 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.

Related

Other skills, from other repositories

triton-ascend-case-reduction-weighted-swiglu

A performance-optimization method for a specialized 3D tensor operation using Weighted SwiGLU during backward computation. It reshapes data and divides work to improve parallel processing while keeping within on-chip memory limits.

wenyi-li/awesome-agent-kernel-skills · 97 tokens

triton-ascend-case-reduction-prod-small

A guide to optimizing small product reductions, which multiply a group of tensor values into one result.

wenyi-li/awesome-agent-kernel-skills · 95 tokens

triton-ascend-case-reduction-sum-large

A guide to optimizing large two-dimensional sum reductions when the non-reduced axis is very large and the reduced axis is medium-sized.

wenyi-li/awesome-agent-kernel-skills · 92 tokens

triton-ascend-case-elemwise-zeros

An optimization for creating small tensors with zeros, arange, or full in Triton Ascend. Triton Ascend is a programming environment for running tensor operations on Ascend hardware.

wenyi-li/awesome-agent-kernel-skills · 67 tokens

agent-platform-rag-engine-management

Manage and query Agent Platform RAG Engine Corpora and retrieve grounded contexts using the Google GenAI SDK. Use when listing RAG corpora or files, inspecting a corpus, retrieving contexts, or generating content grounded in a RAG corpus. Do not use for standard database queries (use SQL/Spanner skills), Google…

google/skills · 85 tokens

agent-platform-model-registry

Agent Platform Model Registry Management. Use when you need to upload, list, describe, update, or delete machine learning models (and their versions) in the Agent Platform Model Registry. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform models.

google/skills · 60 tokens