Nemo Mbridge Perf Moe Dispatcher Selection
by nvidiacf5224d14250Apache-2.03.5K starsListed Oct 8, 2026Updated Oct 8, 2026Repository updated today
Choose the right MoE token dispatcher (`alltoall`, DeepEP, or HybridEP) for the hardware, EP degree, and optimization stage. Summarizes patterns from DSV3, Qwen3, Qwen3-Next, and VLM bring-up work.
Only the file list is public. File contents are available once the skill is installed in a workspace.
| Path | Size | Type |
|---|---|---|
| BENCHMARK.md | 4.3 KB | text/markdown |
| card.yaml | 4 KB | application/yaml |
| evals/evals.json | 1.5 KB | application/json |
| skill-card.md | 3.9 KB | text/markdown |
| SKILL.md | 7.8 KB | text/markdown |
| skill.oms.sig | 4.7 KB | text/plain |
Source and attribution
Source:nvidia/skillsinskills/nemo-mbridge-perf-moe-dispatcher-selectionat commitcf5224d
License: Apache-2.0
Content belongs to its original authors. SourceWeft indexes it from a public repository.
More from nvidia/skills
Tilegym Cutile Python
nvidia
Guides writing, debugging and optimizing cuTile Python GPU kernels, with examples, validation loops and optional multi-agent orchestration.
Tilegym Monkey Patch Kernels To Transformers
nvidia
Integrates TileGym cuTile kernels into Hugging Face transformers models via non-intrusive monkey-patching and auto-generates new kernels.
Tilegym Converting Cutile To Julia
nvidia
Converts cuTile Python GPU kernels to cuTile.jl Julia equivalents, with rules, examples and a static validator.
Tilegym Adding Cutile Kernel
nvidia
Guides adding a new cuTile GPU kernel operator to TileGym, from dispatch registration through tests and benchmarks.
Physicsnemo Discover
nvidia
Guides discovery of PhysicsNeMo model families, datapipes and examples for a SciML task, citing live repo paths.
Nemoclaw User Guide
nvidia
Routes an AI agent to NemoClaw's canonical Markdown docs and docs MCP server to answer setup and operations questions.
More in DevOps & Cloud
Playwright Devops
microsoft
DevOps workflows for Playwright: analyze GitHub Actions failures for the last commit on main and fetch failed job logs.
M5 Onboard
anthropics
Provisions M5Stack ESP32 boards by detecting them on USB, flashing UIFlow 2.0 firmware, and installing a MicroPython app bundle.
Yeet
openai
Stages, commits, pushes, and opens or updates a GitHub pull request in one flow using the GitHub CLI.
Runbook
anthropics
Creates or updates step-by-step operational runbooks for recurring tasks, including troubleshooting, rollback and escalation.
Incident Response
anthropics
Guides an incident response workflow: severity triage, status updates, mitigation tracking, and blameless postmortems.
Deploy Checklist
anthropics
Generates a pre-deployment readiness checklist covering pre-deploy, deploy, post-deploy and rollback triggers.