OpenMOSSOfficial·14 Agent Skills Included

MOSS-VL

Real-time video understanding, GPU kernels, and inference tuning

Runs real-time and offline video understanding with the MOSS-VL open-weight models on continuous streams. Tunes diffusion and LLM inference speed, quantizes models to FP8/NVFP4, and profiles GPU kernels to remove bottlenecks. Guides debugging of CUDA crashes, distributed hangs, and CI regressions without manual trial and error.

410|14

↗

npx skills add OpenMOSS/MOSS-VL --all -g -y

Available:

Tells the agent how to build, test, and debug the FlashAttention kernels, how to run and tune the SGLang diffusion subsystem, and how to write documentation pages following project rules.

All Skills in This Repository (14)

Pure Emerald Level Indicators

📦 In Repo

OpenMOSS

sglang-diffusion-modelopt-quant

Quantize diffusion DiT models with NVIDIA ModelOpt to FP8 or NVFP4 checkpoints for SGLang Diffusion.

sglang-diffusion-benchmark-profile

Analyze SGLang diffusion denoise latency and profile bottlenecks via perf dumps.

sglang-diffusion-add-model

Automate diffusion model integration into SGLang with Hybrid or Modular pipelines.

sglang-diffusion-ako4all-kernel

Optimize SGLang diffusion kernels with AKO4ALL tuning and ncu-guided iteration.

sglang-diffusion-performance

Optimize SGLang diffusion performance settings for a model, GPU, and VRAM budget.

write-sglang-test

Author and register SGLang CI/UT tests using CustomTestCase.

ci-workflow-guide

Document SGLang CI pipeline stage ordering, gating, and fast-fail mechanisms.

add-jit-kernel

Add a lightweight JIT CUDA kernel with host wrapper, tests, and benchmarks to SGLang's jit_kernel module.

debug-cuda-crash

Log CUDA kernel API calls and tensor metadata around crash boundaries in SGLang.

sglang-torch-profiler-analysis

Triage SGLang torch-profiler traces into kernel, overlap, and fuse-pattern tables.

generate-profile

Launch an SGLang server, validate readiness, and generate a Chrome-compatible profiling trace.

sglang-auto-benchmark

Automate SGLang benchmark sweeps with dataset preparation and CSV export.

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install MOSS-VL?▼

Run `npx skills add OpenMOSS/MOSS-VL --all -g -y` in your terminal to install all skills in this suite globally.

What is MOSS-VL used for?▼

MOSS-VL is an open-weight 11B model family for real-time and offline video understanding, supporting interruptible streaming dialogue and long-video question answering.

How to speed up diffusion model inference?▼

The included skills guide you through torch.compile, multi-GPU parallelism, CPU offload, and FP8/NVFP4 quantization with validated benchmark recipes.

How to debug CUDA crashes in SGLang?▼

A dedicated skill shows how to enable kernel API logging, capture crash-safe tensor dumps, and combine them with compute-sanitizer or cuda-gdb.

Do I need GPU experience to use these skills?▼

Basic GPU access is required to run the models, but the skills give step-by-step commands so your agent handles the complex kernel and tuning work.