wenyi-li
- 196 skills
- 0 followers
- 15 hours ago last updated
- ▌ Perf Torch Cuda Graphs · wenyi-li bundleApply CUDA Graphs to PyTorch workloads — API selection (torch.compile, PyTorch make_graphed_callables, TE make_graphed_callables, MCore CudaGraphManager, FullCudaGraphWrapper, manual torch.cuda.graph), code compatibility, capture workflows, dynamic pattern handling, and troubleshooting. Triggers: CUDA graph, torch.cuda.graph, make_graphed_callables, reduce-overhead, graph capture, graph replay, kernel launch overhead, CudaGraphManager, FullCudaGraphWrapper, full-iteration graph, stream capture.
- ▌ Rocm Profiler Analysis · wenyi-li bundleAnalyze SGLang and vLLM profiler traces on AMD ROCm systems, especially MI355X/gfx950 nodes. Adapted from the SGLang torch-profiler workflow: triage kernel breakdown, overlap headroom, and fuse opportunities, then write structured artifacts that can be attached to amdpilot experiments, trials, and dashboard views. Use when a run needs profiling, when an optimization trial should produce machine-readable profiling artifacts, or when the user asks why a ROCm workload is slow.
- ▌ Tilelang Ascend Kernel · wenyi-liUse when the user wants to add, modify, debug, or review an xLLM TileLang Ascend kernel or specialization, including Python kernel definitions, generated Ascend-C source, runtime wrapper dispatch, TileLang CMake wiring, and NPU tests.
- ▌ Tilelang Cuda Patterns · wenyi-liTileLang CUDA 核心编程模式（逐元素、归约、矩阵乘法、GEMV）的标准实现范式和代码模板。适用于需要快速确定算子属于哪种编程模式、或需要了解 TileLang 各模式基本代码结构的内核代码生成场景
- ▌
- ▌
- ▌ Amd Kernel Optimization · wenyi-li bundleOptimize inference latency and throughput of PyTorch models on AMD GPUs (MI250/MI300/MI350) with ROCm. Use when profiling and optimizing GEMM, attention, elementwise ops, torch.compile, CUDAGraphs, or Triton kernels on AMD hardware. Covers the full optimize cycle: benchmark → profile → analyze → implement → verify. Also covers benchmarking methodology and common pitfalls that waste time.
- ▌ Flydsl Kernel Authoring · wenyi-li bundleFlyDSL is a Python DSL with MLIR-native backend for authoring custom AMD GPU kernels with explicit layout algebra (pre-installed at /opt/FlyDSL on images tagged *-flydsl:*). Use this skill when profiling identifies a hot per-row reduction (RMSNorm / LayerNorm / softmax), a fused elementwise chain (norm + residual add, activation + multiplier), or an unusual-shape grouped GEMM that the standard AMD backends (Triton / aiter / CK / hipBLASLt / TransformerEngine) don't serve well. Essential for any workload where Python/config/Triton-tuning gains have plateaued and the profile shows a custom kernel opportunity. Covers the `/opt/FlyDSL` availability check, the integration playbook (dispatcher + direct site-packages edit + autograd-safe output handling), kernel authoring patterns (elementwise via layout API, block reductions via wave_reduce_add, fused dx+dw designs, MFMA GEMM preshuffle), torchrun gotchas, and the critical rule that custom kernels typically only win end-to-end when stacked with `torch.compile(mode=
- ▌ Hip Kernel Optimization · wenyi-liThis skill should be used when writing or tuning HIP kernels on AMD/NVIDIA GPUs, covering memory coalescing, shared-memory tiling, bank conflict avoidance, warp primitives, occupancy, vectorization, async ops, loop unrolling, and profiling.
- ▌ Perf Workload Profiling · wenyi-li bundleCode instrumentation for timing workloads. Two scenarios: (1) Training loop — inject manual timing to report per-iteration latency, throughput (samples/sec), and data load time. (2) Standalone kernel/op — write CUDA event timing code with warmup, per-iteration statistics, and anti-pattern avoidance. Also covers NVTX annotation for labeling profiler timelines. NOT for: running or analyzing profiler tools (nsys, ncu, Nsight Systems, Nsight Compute), writing kernels (Triton, CuTe, CUDA), applying optimizations (CUDA Graphs, gradient checkpointing, fusion), or interpreting roofline/SOL% metrics. Triggers: "measure throughput", "benchmark this function", "time my training loop", "samples per second", "NVTX annotate", "instrument my dataloader", "data load time", "kernel timing", "how do I time".
- ▌ Optimizing Tilelang Programs · wenyi-li bundleHow to optimize TileLang GPU kernels for better performance. Use this skill whenever the user has a working TileLang kernel that is too slow, wants to tune tile sizes or pipeline stages, needs to improve TFLOPS or bandwidth utilization, wants to use the AutoTuner, or asks questions like "how do I make my kernel faster", "what tile sizes should I use", "how do I autotune", or "why is my kernel slower than cuBLAS". Also trigger when the user mentions block sizes, num_stages, threads per block, shared memory pressure, occupancy, swizzle, vectorization, split-K, or any performance tuning in TileLang context. Even for vague requests like "improve performance" or "speed up this kernel" when TileLang is involved.
- ▌ Perf Nsight Compute Analysis · wenyi-li bundleAnalyze ncu (NVIDIA Nsight Compute) profiling output: SOL% bottleneck classification, roofline analysis, occupancy diagnosis, memory hierarchy analysis, warp stall analysis, metric interpretation, and programmatic .ncu-rep report analysis. NOT for kernel writing or code generation, Nsight Systems (nsys), host-side profiling, or system-level profiling.
- ▌ Sglang Diffusion Performance · wenyi-liUse when choosing the fastest SGLang Diffusion flags for a model, GPU, and VRAM budget.
- ▌
- ▌
- ▌
- ▌ Gpu Architecture Fundamentals · wenyi-liThis skill should be used when reasoning about GPU architecture fundamentals to guide kernel optimization choices such as memory hierarchy usage, execution model mapping, block sizing, and latency-aware tuning across HIP, Triton, and PyTorch.
- ▌
- ▌
- ▌ Tilelang Cuda Synchronization · wenyi-liTileLang CUDA 同步规范，包括 T.sync_threads() 使用规则、线程安全最佳实践和死锁预防策略。适用于编写涉及共享内存访问、多线程协作、或需要避免同步死锁的 TileLang 内核代码生成场景
- ▌
- ▌
- ▌ Mi300 Hip Programming Insights · wenyi-li bundleCDNA3/MI300 HIP programming insights—chiplet/cache model, Infinity Cache, memory coherency, matrix cores, sparsity, and best practices.
- ▌ Rocprofv3 Profiler · wenyi-li bundleProfile AMD GPU kernels using rocprofv3 and analyze performance bottlenecks. Use when the user wants to profile HIP/ROCm kernels, identify GPU performance issues, analyze hardware counters, or understand why a kernel is slow on AMD GPUs (MI100, MI200, MI300 series). Provides wrapper scripts for rocprofv3 execution and automated parsing of profiler output into structured, agent-friendly JSON with bottleneck classification.
- ▌ Cuda Skill · wenyi-li bundleQuery NVIDIA PTX ISA 9.1, CUDA Runtime API 13.1, Driver API 13.1, Programming Guide v13.1, Best Practices Guide, Nsight Compute, Nsight Systems local documentation. Debug and optimize GPU kernels with nsys/ncu/compute-sanitizer workflows. Use when writing, debugging, or optimizing CUDA code, GPU kernels, PTX instructions, inline PTX, TensorCore operations (WMMA, WGMMA, TMA, tcgen05), or when the user mentions CUDA API functions, error codes, device properties, memory management, profiling, GPU performance, compute capabilities, CUDA Graphs, Cooperative Groups, Unified Memory, dynamic parallelism, CUDA programming model concepts, bank conflicts, shared memory optimization, warp divergence, memory coalescing, occupancy tuning, register pressure, L2 cache control, async copy, mbarrier, thread block clusters, or CUDA architecture questions (Ampere sm_80, Hopper sm_90, Blackwell sm_100).
- ▌ Tilelang To Flydsl · wenyi-li bundlePort a kernel written in TileLang (the `@T.prim_func` / `with T.Kernel(...)` DSL used by TileKernels and other tile-ai projects) into an equivalent FlyDSL kernel (`@flyc.kernel` / `@flyc.jit` with explicit layout algebra, copy atoms, MMA atoms, and SmemAllocator). Use whenever the task is to rewrite, translate, or migrate a TileLang `@T.prim_func` body into FlyDSL, including converting individual operators in the TileKernels test suite while preserving the Python wrapper signatures so the existing pytest cases keep validating the new implementation.
- ▌ Triton Cuda Basics · wenyi-liTriton CUDA 编程基础，包括核心概念（program_id、block、grid）、内核函数结构、装饰器用法和标准代码模式。适用于使用 Triton CUDA、需要了解基本语法结构的任意 CUDA 内核代码生成场景
- ▌ Triton Cuda Matmul · wenyi-li矩阵乘法算子(matmul/bmm/linear)优化策略，包括分块 Tiling、共享内存缓存、Tensor Core 利用和大矩阵处理技巧。适用于实现 GEMM、批量矩阵乘、全连接层等矩阵运算的 CUDA 内核代码生成场景
- ▌ Triton Cuda Memory · wenyi-liCUDA GPU 内存访问优化策略，包括共享内存利用、合并访存、Bank Conflict 避免和数据布局优化技巧。适用于内存带宽受限、需要优化全局内存访问效率、或处理大规模数据的 CUDA 内核性能优化场景
- ▌
- ▌ Cuda Code Generator · wenyi-li bundleGenerates optimized CUDA kernel code based on performance analysis reports or algorithm type. Reads NCU analysis reports (e.g. *_analysis.md) and optionally existing kernel code, then produces high-quality compilable .cu files with applied optimizations. Use when the user provides an NCU analysis report or requests CUDA kernel generation, optimization, or implementation of techniques like Shared Memory Tiling, vectorized loads, bank conflict elimination, or double buffering. Does not handle compilation, execution, or profiling.
- ▌ Kernel Cute Writing · wenyi-li bundleWrite and implement GPU kernels using NVIDIA CuTe DSL (CUTLASS 4.x Python API) — NOT for Triton, CUDA C++, or conceptual explanations. Trigger only when the user wants to write or implement a kernel, not when asking questions about CuTe DSL concepts or layouts. CuTe DSL uses cute.jit/cute.kernel decorators and cutlass.cute imports. Covers element-wise kernels, GEMM patterns, reductions, memory hierarchy (global/shared/register/TMA), MMA tensor core operations, software pipelining, and framework integration.
- ▌ Mi300 Hip Vs Nvidia · wenyi-li bundleMI300 HIP programming differences vs NVIDIA—wavefront vs warp, memory hierarchy, MFMA usage, occupancy, and profiling pitfalls.
- ▌
- ▌
- ▌
- ▌ Triton Kernel Reflection Prompts · wenyi-li bundleReflection/self-critique prompts for reviewing and fixing AMD-targeted Triton kernels after generation or test failures.
- ▌ Torch Profiling Tilelang Programs · wenyi-li bundleUse torch.profiler as a lightweight, always-available alternative to Nsight Compute/Systems for profiling, debugging, and optimizing TileLang GPU kernels. Trigger this skill whenever the user wants to profile a TileLang kernel without installing ncu/nsys, get an operator/kernel breakdown of a forward pass that includes TileLang custom ops, hunt launch-overhead in a multi-kernel pipeline, classify whether a TileLang kernel is IO-bound / CUDA-core bound / Tensor-core bound, generate a Chrome trace, profile GPU memory allocations, or compare a TileLang kernel against a torch/cuBLAS reference inside one trace. Also trigger on phrases like "profile this kernel", "why is this slow", "how do I see the kernel timeline", "lightweight profiling", "torch profiler", "chrome trace", or "perfetto", even if the user does not mention TileLang explicitly but TileLang code is on the table.
- ▌
- ▌ Sglang Diffusion Benchmark Profile · wenyi-li bundleUse when benchmarking denoise latency or profiling a diffusion bottleneck in SGLang.
- ▌ Tilegym Improve Cutile Kernel Perf · wenyi-li bundleIteratively optimize cuTile kernel performance through systematic profiling, bottleneck analysis, IR comparison, and targeted tuning. Covers tile sizes, occupancy, autotune configs, TMA, latency hints, persistent scheduling, num_ctas, flush_to_zero, and IR-level debugging. Use when asked to "optimize cutile kernel", "improve kernel perf", "tune cutile performance", "make kernel faster", or iteratively benchmark and refine a cuTile GPU kernel in the TileGym project.
- ▌ Triton Hip Reference Kernel Search · wenyi-li bundleSearch and adapt Triton/HIP kernel patterns from a corpus to optimize AMD GPUs; use to find similar ops and reuse tiling/occupancy strategies.
- ▌
- ▌
- ▌ Write Triton Fused Add Rmsnorm Kernel · wenyi-li bundleSkill: Write a Triton Fused Add+RMSNorm Kernel
- ▌ Write Vllm Custom Op Integration Plan · wenyi-li bundleSkill: Write a vLLM Custom Op Integration Plan
- ▌ Perf Nsight Systems · wenyi-li bundleNsight Systems (nsys) CLI for system-level timeline profiling. Use when the user wants to run nsys profile, analyze .nsys-rep reports, use nsys stats/analyze/recipe commands, diagnose GPU idle time from timeline traces, or profile distributed training with NCCL overlap analysis. NOT for kernel-level metrics like SOL%, occupancy, or roofline (use perf-nsight-compute-analysis for ncu). NOT for writing or generating kernels. NOT for applying optimizations like CUDA Graphs.
- ▌
- ▌
- ▌ Tilelang Cuda Memory · wenyi-liTileLang CUDA 内存访问优化策略，包括 T.alloc_shared/fragment 使用、数据布局优化、合并访存和 Bank Conflict 避免技巧。适用于内存带宽受限、需要优化数据搬运效率的 TileLang 内核性能优化场景
- ▌ Tle Developer Flagos · wenyi-li bundleSelf-contained orchestration skill for writing high-performance TLE kernels and shipping TLE feature changes with reproducible validation. Use when the user wants to write/optimize TLE kernels, implement TLE API/verifier/lowering features, or debug TLE correctness/performance issues. Trigger on phrases like "write a TLE kernel", "optimize TLE operator", and "debug TLE local_ptr".
- ▌ Triton Ascend Basics · wenyi-liTriton Ascend 编程基础，包括核心概念（program_id、block、grid）、内核函数结构、装饰器用法和标准代码模式。适用使用 Triton Ascend、需要了解基本语法结构的任意内核代码生成场景
- ▌ Triton Ascend Matmul · wenyi-li适用于矩阵乘法(matmul)类算子的优化指南。当算子的核心计算涉及二维或更高维的矩阵乘法时应选择此指南，典型算子包括：matmul, mm, bmm, linear, gemm, outer_product, einsum(含矩阵乘), conv(转为矩阵乘实现)等。涵盖 Cube Core 使用、分块(tiling)策略、Swizzle 优化、大 K 维处理等关键技巧。不适用于纯逐元素运算或纯归约运算。对于 attention 机制中的 QK^T 和 score*V 矩阵乘，若算子整体是注意力计算，应优先选择 attention 指南。
- ▌ Triton Ascend Memory · wenyi-liAscend NPU 内存访问优化策略，包括 UB（统一缓冲区）利用、数据布局优化、合并访存和预取技巧。适用于内存带宽受限、需要优化数据搬运效率、或处理大规模数据的内核代码性能优化场景
- ▌ Triton Cuda Patterns · wenyi-liTriton CUDA 三大核心编程模式（向量/逐元素、归约、矩阵乘法）的标准实现范式和代码模板。适用于需要快速确定算子属于哪种编程模式、或需要了解各模式基本代码结构的 CUDA 内核代码生成场景
- ▌
- ▌ Kernel Triton Writing · wenyi-li bundleONLY 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).
- ▌ Triton Cuda Attention · wenyi-liAttention 算子的 Triton-CUDA 实现指南。包含经过验证的 Flash Attention 完整示例、各变体（Causal/GQA/MQA/RoPE）的差异改法、在线 Softmax 算法和常见错误
- ▌ Triton Cuda Debugging · wenyi-liTriton CUDA 调试排查清单和常见错误速查表，包括编译错误、运行时错误、精度问题和性能问题的诊断方法。适用于 CUDA 内核代码出现错误需要定位原因、或需要验证代码正确性的调试场景
- ▌ Kernel Designer · wenyi-li bundle算子算法草图设计 Skill — 负责根据任务需求设计高质量的算法草图（sketch），提供伪代码形式的算法方案、优化建议和实现策略。 支持多种 DSL：triton_cuda、triton_ascend、cpp、cuda_c、tilelang_cuda、pypto。 支持 Hint 模式（参数空间配置）。
- ▌ Kernel Verifier · wenyi-li bundle算子代码验证 Skill — 静态代码检查 + 精度对比验证。 包含两阶段验证：先做零成本静态检查（语法、编译、import、DSL 合规性）， 通过后再对比框架实现与生成实现的输出一致性。 支持多框架（torch / mindspore）、多后端（cuda / ascend / cpu）。
- ▌ Kernel Workflow · wenyi-li bundleAI Kernel 算子生成与优化工作流程。当用户需要生成、验证或优化 kernel 算子时使用此 Skill。支持 Triton、CUDA C、C++、TileLang等多种后端 DSL。
- ▌ Rocprof Compute · wenyi-li bundleThis skill should be used when profiling AMD GPU kernels with rocprof-compute to collect metrics, roofline data, and analyze bottlenecks for HIP kernels.
- ▌ Triton Cuda API · wenyi-liTriton CUDA API 完整参考手册，包括 tl.load/store、tl.reduce、tl.dot、tl.atomic 等核心函数的签名、参数和使用示例。适用于需要查阅具体 API 用法、了解函数参数含义的任意 Triton CUDA 内核代码生成场景
- ▌ Aiter Reflection · wenyi-liThis skill should be used when optimizing AMD GPU kernels on MI300 using the aiter project, including running op tests, benchmarking, iterating on kernel changes, and recording results in the kernel experiment database.
- ▌ Kernelgen Flagos · wenyi-li bundleUnified GPU kernel operator generation and optimization skill. Automatically detects the target repository type (FlagGems, vLLM, or general Python/Triton) and dispatches to the appropriate specialized sub-skill. Includes operator generation, MCP-based iterative optimization, and feedback submission sub-skills. Use this skill when the user wants to generate or optimize a GPU kernel operator, create a Triton kernel, or says things like "generate an operator", "create a kernel for X", "optimize triton kernel", or "/kernelgen-flagos".
- ▌ Ncu Rep Analyzer · wenyi-liProfiles a CUDA kernel with NCU and analyzes the resulting .ncu-rep report to diagnose performance bottlenecks and generate optimization recommendations. Use when the user provides a .cu file or a .ncu-rep file and asks for performance analysis, NCU profiling, or bottleneck diagnosis. If given a .cu file, runs NCU via benchmark.py to produce a .ncu-rep, then imports it with `ncu --import` to extract metrics (SM throughput, DRAM/L1 bandwidth, occupancy), classifies the bottleneck (DRAM-bound, compute-bound, latency-bound, etc.), and saves a structured *_analysis.md report alongside the .ncu-rep file.
- ▌ Ncu Report Skill · wenyi-li bundleProfile CUDA kernels with Nsight Compute on B200 / sm_100. Use when the user asks to profile a kernel, analyze its performance, diagnose bottlenecks, read an ncu report, or write an optimization plan — including variants in Chinese ("profile 一下", "为什么慢", "ncu 报告").
- ▌ Perf Test Flagos · wenyi-li bundleRun accuracy benchmarks (FlagEval, when available) and performance benchmarks (vllm bench serve) against a served model. Covers 5 workload profiles: short/long prefill x short/long decode + high concurrency. Collects throughput, latency, TTFT, TPOT metrics.
- ▌ Rocm Crash Debug · wenyi-li bundleDebug ROCm/HIP kernel crashes in SGLang and vLLM on AMD GPUs (MI300X/MI325X/MI355X). Adapts SGLang's @debug_kernel_api kernel boundary logging to ROCm: captures input tensors before crash, tracks shapes/dtypes/values, dumps crash artifacts for offline analysis. Integrates with amdpilot executor failure_reason field and dashboard trajectory viewer. Triggered by: CUDA/HIP errors, illegal memory access, device-side assert, OOM kills, signal 137/139, NaN/Inf in outputs, "debug crash", "why did the trial fail".
- ▌
- ▌ Perf Optimization · wenyi-liPerformance optimization coordination playbook. Contains specialist routing table, TileIR two-step pipeline, kernel generation specialist selection, prioritization criteria, and safe modification workflow. Use when the user asks to apply optimizations, write kernels, or improve performance. Covers both user-specified optimization and autopilot-driven iterative optimization.
- ▌ Tilelang Cuda API · wenyi-liTileLang CUDA API 完整参考手册，适用于需要查阅具体 API 用法、了解函数参数含义的任意 TileLang CUDA 内核代码生成场景
- ▌
- ▌ Ncu Cuda Profiling · wenyi-liAutomated NCU (Nsight Compute) profiling workflow with full metrics collection and persistent storage
- ▌ Perf Host Analysis · wenyi-li bundleAnalyze host/CPU overhead in TensorRT-LLM inference from nsys traces. Detect whether host overhead is the bottleneck using GPU idle ratio, host prep exposed ratio, and per-phase evidence. For regressions, isolate forward steps via allreduce/NVTX patterns, compare host operation breakdowns across versions, and identify scheduling or request-management overhead. Supports optional inter-kernel gap, eager-vs-graph, pattern mapping, and multi-rank straggler drill-down. Use standalone or within perf-analysis. Triggers: host overhead, inter-step gap, scheduling overhead, forward step isolation, nsys iteration analysis, NVTX breakdown, request management overhead, GPU idle, host bottleneck, host prep exposed, inter-kernel gap, bubble analysis, graph coverage, eager kernel, rank imbalance, straggler detection.
- ▌ Add Jit Kernel · wenyi-liStep-by-step tutorial for adding a new lightweight JIT CUDA kernel to sglang's jit_kernel module
- ▌ Auto Benchmark Rocm · wenyi-li bundleRun AI-driven benchmark searches on AMD ROCm with tiered server-flag sweeps for vLLM/SGLang, canonical dataset preparation, SLA or fixed-QPS benchmarking, CSV export, and resume. Adapted from SGLang auto-benchmark for MI355X (gfx950) / MI300X (gfx942) on ROCm 7.x. Use when the user wants an automated benchmark workflow on AMD GPUs rather than a one-off bench_serving command. Integrates with amdpilot executor task_spec_json for resource-aware launch.
- ▌ Cuda Knowledge · wenyi-li bundleCUDA kernel development, debugging, performance optimization, linear algebra, and multi-GPU communication for Claude Code. Use when writing, debugging, or optimizing CUDA code, GPU kernels, parallel algorithms, or CUDA library calls. Covers cuBLAS/cuBLASLt GEMM operations, CUDA Math API (half, bfloat16, FP8, FP6, FP4), NCCL multi-GPU collectives, non-interactive profiling with nsys/ncu, debugging with cuda-gdb/compute-sanitizer, binary inspection with cuobjdump, and performance analysis workflows. Triggers on CUDA, GPU programming, kernel optimization, nsys, ncu, cuda-gdb, compute-sanitizer, PTX, GPU profiling, parallel performance, cuBLAS, cublasLtMatmul, GEMM, GemmEx, FP8, bfloat16, half precision, __half, __nv_bfloat16, cublasGemmEx, cublasGemmStridedBatchedEx, NCCL, ncclAllReduce, ncclReduceScatter, ncclAllGather, ncclCommInitRank, tensor parallel, pipeline parallel, all-reduce, vLLM CUDA kernels.
- ▌ Cuda Optimizer · wenyi-liOrchestrates a full profiling-driven CUDA kernel optimization loop (write → validate → profile → analyze → optimize) until performance converges or no further gains are possible. Capabilities include generating reference implementations, writing initial kernels via cuda-code-generator, running correctness validation and benchmarks via kernel-benchmarker, profiling and analyzing NCU reports via ncu-rep-analyzer, and applying targeted optimizations via cuda-code-generator. Use when the user wants to optimize a .cu file or CUDA kernel, improve GPU kernel performance, run a CUDA optimization workflow.
- ▌ Kernel Creator · wenyi-liGuides writing and refactoring GPU kernels across CUDA and DSLs. Invoke when implementing kernel files that must stay importable, independently testable, profileable, and optimizable.
- ▌ Triton Kernels · wenyi-li bundleWrite optimized Triton GPU kernels for deep learning operations. Covers the full spectrum from basic vector ops to Flash Attention, persistent matmul, fused normalization, quantized GEMM, and memory-efficient patterns.
- ▌
- ▌ Amd Rocm Porting · wenyi-li bundlePort NVIDIA CUDA codebases to AMD ROCm GPUs. Use when making PyTorch models run on AMD GPUs, replacing NVIDIA-specific libraries with AMD equivalents, fixing ROCm build/runtime failures, or porting C/C++ CUDA kernels to HIP. Also covers dependency debugging and environment setup on ROCm Docker images.
- ▌ Kernel Generator · wenyi-li bundle算子内核代码生成 Skill — 负责算子实现的全部智力工作：方案讨论、代码生成、基于反馈修改。 支持多种 DSL：triton_cuda、triton_ascend、cpp、cuda_c、tilelang_cuda、pypto。
- ▌ Cuda Samples · wenyi-li bundleCurated index of NVIDIA official CUDA Samples with code patterns, GitHub permalinks, and key snippets. Use when looking for working CUDA code examples, GPU kernel patterns (reduction, scan, GEMM, transpose, histogram), Tensor Core WMMA usage, CUDA Graphs API, cuBLAS/cuFFT/cuSPARSE library integration, stream/event async patterns, multi-GPU P2P/IPC, or framework interop (PyTorch/TensorFlow custom kernels). Covers 50+ curated samples organized by topic. Triggers on CUDA sample, code example, kernel pattern, GEMM example, reduction example, CUDA graph example, cuBLAS example, Tensor Core WMMA, how to write CUDA kernel, CUDA code reference.
- ▌ Sglang Skill · wenyi-li bundleDevelop, debug, and optimize SGLang LLM serving engine. Use when the user mentions SGLang, sglang, srt, sgl-kernel, LLM serving, model inference, KV cache, attention backend, FlashInfer backend, MLA, MoE routing, MoE dispatch, expert parallelism SGLang, speculative decoding, disaggregated serving, TP/PP/EP, radix cache, continuous batching, chunked prefill, CUDA graph SGLang, model loading, quantization FP8/GPTQ/AWQ, JIT kernel, triton kernel SGLang, DeepSeek serving, EPLB (expert load balancing), HiCache, launch_server, sglang Engine API, LoRA inference, torch.compile SGLang, or asks about serving LLMs with SGLang. Also use when the user wants to add a new model to SGLang, add a new attention backend, debug SGLang serving issues, or optimize SGLang throughput/latency.
- ▌ Triton Skill · wenyi-li bundleWrite, debug, and optimize Triton and Gluon GPU kernels using local source code, tutorials, and kernel references. Use when the user mentions Triton, Gluon, tl.load, tl.store, tl.dot, tl.dot_scaled, triton.jit, gluon.jit, wgmma, tcgen05, TMA, tensor descriptor, persistent kernel, warp specialization, fused attention, matmul kernel, kernel fusion, tl.program_id, triton autotune, MXFP, FP8, FP4, NVFP4, block-scaled matmul, SwiGLU, top-k, triton_kernels, roofline analysis, Triton IR, TritonGPU dialect, MLIR Triton, PDL (programmatic dependent launch), cluster launch control, or asks about writing GPU kernels in Python. Also use when the user wants to understand Triton compiler internals, debug Triton kernel correctness, profile Triton kernel performance, or convert CUDA kernels to Triton.
- ▌
- ▌ Cutlass Skill · wenyi-li bundleWrite, debug, and optimize CUTLASS and CuTeDSL GPU kernels using local source code, examples, and header references. Use when the user mentions CUTLASS, CuTe, CuTeDSL, cute::Layout, cute::Tensor, TiledMMA, TiledCopy, CollectiveMainloop, CollectiveEpilogue, GEMM kernel, grouped GEMM, sparse GEMM, flash attention CUTLASS, blackwell GEMM, hopper GEMM, FP8 GEMM, FP4 GEMM, blockwise scaling, MoE GEMM, StreamK, warp specialization CUTLASS, TMA CUTLASS, epilogue fusion, EVT (Epilogue Visitor Tree), pycute, Layout algebra, Swizzle pattern, GemmUniversal, KernelSchedule, EpilogueSchedule, CUTLASS collective builder, CUTLASS pipeline, or asks about writing high-performance CUDA kernels with CUTLASS/CuTe templates. Also use when the user wants to understand CUTLASS source code structure, compile CUTLASS examples, or debug CUTLASS template errors.
- ▌ Perf Analysis · wenyi-liPerformance analysis coordination workflow. Guides profiling delegation, bottleneck classification (compute/memory/launch/communication/sync), and structured report generation. Use when the user asks to analyze performance, profile a workload, check MFU/SOL, or diagnose bottlenecks.
- ▌ Tilelang Wiki · wenyi-li bundleA local, code-grounded TileLang reference skill for DSL semantics, kernel authoring, API lookup, tuning, debugging, examples, and compiler/runtime behavior. Use whenever you need to write, modify, explain, optimize, validate, or troubleshoot TileLang programming work.
- ▌
- ▌ Kernelwiki · wenyi-liUse when the user asks about optimizing NVIDIA Blackwell (SM100, B200) or Hopper (SM90, H100) GPU kernels — tcgen05/TMEM/CLC/NVFP4/2-SM cooperative, warp specialization, FlashAttention-4, DeepGEMM, FlashMLA, MoE, grouped GEMM, CuTe-DSL/PTX/Triton on Blackwell, or wants concrete PR references from CUTLASS/SGLang/vLLM/FlashInfer/PyTorch. Do NOT use for generic CUDA Q&A that is not Blackwell/Hopper-specific, host-side framework integration, or distributed systems (DeepEP/EPLB/DualPipe).
- ▌ Xpu Kernels · wenyi-li bundleProvides guidance for writing, optimizing, and benchmarking Triton kernels for Intel XPU GPUs (Battlemage/Arc Pro B50) using the Xe-Forge optimization framework. Includes an LLM-driven trial-loop workflow (analyze, validate, benchmark, profile, finalize), XPU-specific patterns (tensor descriptors, GRF mode, tile swizzling), KernelBench fused kernels, and Flash Attention.
- ▌ Cuda Kernels · wenyi-li bundleProvides guidance for writing and benchmarking optimized CUDA kernels for NVIDIA GPUs (H100, A100, T4) targeting HuggingFace diffusers and transformers libraries. Supports models like LTX-Video, Stable Diffusion, LLaMA, Mistral, and Qwen. Includes integration with HuggingFace Kernels Hub (get_kernel) for loading pre-compiled kernels. Includes benchmarking scripts to compare kernel performance against baseline implementations.