catlass-dsl-generator

AISS-TilingSolver：Ascend C 算子 Tiling 参数自动求解工具（MatMul / Vector）。下载安装、构造 JSON 输入、运行求解、结果解读与故障排查。

94 skills

Ascend C 代码审查 Skills（2 个）。代码审查规范和 API 文档检索。

94 skills

Ascend C Kernel 直调算子开发 Skills（40+ 个）。API 检索、Tiling 设计、精度调试、性能采集、环境检查、代码模板等。

94 skills

Ascend C 算子性能进化优化共享 Skills（2 个）。NPU 架构与芯片规格查询、性能采集与分析。

94 skills

AscendC 算子性能调优 Skills（6 个）。性能仿真、Profiling 采集、Tiling 建模与流水分析、性能优化最佳实践、MC² 最佳实践、API 文档检索。

94 skills

Ascend C 注册调用算子开发 Skills（21 个）。API 检索、Tiling 设计、精度调试、性能采集、环境检查、代码模板、测试设计、UT 开发等。

94 skills

MindSpeed-LLM skill set for Huawei Ascend NPU, covering the full training pipeline: environment setup, data preprocessing, weight conversion, distributed training, and end-to-end orchestration. Use when deploying MindSpeed-LLM training on Ascend NPU.

191 skills

Hardware backend guidance for CUDA, ROCm, NPU, and XPU.

Control-plane plugin for tracking rings: recursive units of work, each a problem to settle rather than a session. A ring has a Brief frozen at open, a Trace of local working memory discarded at close, and a Delta typed by effect (ESTABLISHED, INVALIDATED, SURFACED, DELIVERABLE) that crosses exactly one turn to its parent. Ring state is orthogonal to what carries it: the four vehicles are inline, subagent, session and resumed. For vehicle session, ring layers on top of the team plugin as its vehicle — team keeps circulating mandates, reports and relay packets between named agents unchanged, while ring tracks which problem is open, who is carrying it, and what crosses back up. Ships four skills (open, close, rings, sync) for descending into and ascending out of a ring, listing/graphing the live tree, and persisting a ring's state so it survives a /clear, an on-disk writer token (carrier:) taken over by resuming a parked or active ring, exclusive create:/writes: write-right tracking with prefix-inclusion collision detection and a lineage exception for parent-child delegation, and a closing sweep that scans disk for files born or touched outside declared claims.

4 skills

在 Ascend 950 / DAV_3510 上用户明确指定 Tensor API 时的实现路线权威源（硬约束，不由 Blaze 优先选型改写），Tensor API 为主、C API 为辅，覆盖 API 分工、Layout、内存预算、工具链、精度与性能。

109 skills

PyTorch algorithm -> Ascend PTO-ISA kernel generation pipeline (decompose, generate, validate, benchmark, optimize, fuse).

6 skills · 3 agents · 1 MCP server

PyTorch to Ascend NPU migration analyzer and optimizer. Analyze models for Ascend compatibility and generate fully optimized migrated code with ALL performance optimizations (fused attention, torch.compile, data transfer optimization, etc.).

6 skills · 3 commands