moirai-internal
Official@moirai-internal
Optimizing high-performance diffusion inference through custom CUDA, Triton kernel integration, and rigorous regression testing within the SGLang framework.
Agent Skills by moirai-internal
Showing 7 vetted skills indexed across 1 GitHub repositories.
moirai-internaladd-new-diffusion-model
Guide adding diffusion models to the SGLang framework with pipeline configuration.
moirai-internaldiffusion-optimal-perf
Optimize SGLang-Diffusion inference with CLI flags and environment variables.
moirai-internaldiffusion-kernel
Integrate optimized CUDA and Triton kernels for diffusion model inference.
moirai-internalwrite-sglang-test
Write SGLang CI and unit tests following project conventions.
moirai-internaladd-jit-kernel
Guide users through adding a lightweight JIT CUDA kernel to SGLang's jit_kernel module.
moirai-internaladd-sgl-kernel
Adds custom CUDA/C++ kernels to the sgl-kernel library with PyTorch extension integration and CMake build system support.
moirai-internalSGLang Bisect CI Regression
Bisect SGLang CI logs to identify regression-introducing commits.
Frequently Asked Questions About moirai-internal
FAQPage SchemaWhat specific tasks can be performed using these diffusion optimization skills?▼
These skills enable the integration of custom CUDA and Triton kernels into the SGLang framework, configuration of high-performance inference pipelines, and systematic identification of performance regressions through CI log bisection. Users can extend existing model support and optimize execution environments for diffusion-based architectures.
Which technical personas are the primary target for these capabilities?▼
The primary target personas are machine learning engineers, kernel developers, and performance optimization specialists working on high-throughput generative model deployment. These individuals typically focus on low-level hardware acceleration, framework-level integration, and maintaining stability in complex distributed inference environments.
What are the prerequisites for implementing custom kernels within this framework?▼
Implementation requires proficiency in C++, CUDA, and PyTorch extension development, alongside familiarity with the CMake build system. Users must understand the SGLang architecture and the specific requirements for integrating JIT-compiled kernels into the existing library structure to ensure compatibility and performance gains.