
Pathways workload
SkillDiagnoses failing, crashing or hung Pathways on Cloud JAX workloads running on GKE TPUs
- Open source
- 24.4K GitHub stars
- Cloud and infrastructure

Diagnoses failing, crashing or hung Pathways on Cloud JAX workloads running on GKE TPUs

Lets an agent run Google Cloud Workload Manager evaluations and review their results

Deploy Expo web apps and API routes to EAS Hosting and author Cloudflare-ready routes

Plans and applies reversible changes to free Jetson memory by going headless

Recommends an LLM serving runtime and memory flags for NVIDIA Jetson devices

Guides an agent through launching a vLLM or SGLang OpenAI-compatible server on Jetson

Sets up NeMo AutoModel job launches on interactive, Slurm and SkyPilot runs

Guides installing, troubleshooting and connecting the NVIDIA Auto Ontology runtime

Recommends NVSHMEM NIC-to-PE mappings and environment exports from GPU and RDMA topology

Sets up and validates NVIDIA physical AI infrastructure for synthetic data workflows

Audits GPU quota fairness in a Run:ai cluster, finding starved and over-borrowing projects

Diagnoses why a Run:ai workload failed, crashed or stopped, with cited evidence

Checks a Run:ai cluster's node health and scheduling, with a per-pool verdict

Finds Run:ai GPUs held by projects or departments that sit idle or underused

Summarizes Run:ai workload health across a scoped set or for one workload, read-only

Moves TAO job data to and from compute containers, choosing the storage tier first

Gates TAO model launches behind preflight checks, image confirmation and a review step

Runs TAO model checkpoints as local inference microservices with start, query and stop

Runs TAO training, evaluation and inference containers on Brev GPU instances

Lets an agent submit, monitor and cancel TAO GPU training jobs on Kubernetes clusters

Runs TAO GPU jobs on a remote SLURM cluster over SSH with sbatch and Pyxis containers

Checks and installs the NVIDIA driver, CUDA and container toolkit for TAO GPU hosts

Launches the AutoMagicCalib microservice and web UI from NGC images with Docker Compose

Audits a Linux host and Kubernetes cluster for Cosmos Data Factory controller readiness

Deploys and verifies an Airflow controller for CDF orchestration on Kubernetes

Guides running DOCA-linked binaries directly on BlueField hosts and DPU Arm cores

Guides bringing up a BlueField-3 DPU over RShim: BFB push, TMFIFO link and recovery

Guides an agent to operate NVIDIA DOCA Management Service on BlueField and ConnectX

Guides deploying and configuring NVIDIA DOCA Telemetry Service on BlueField, host or DPF

Guides agents through DOCA Ethernet RX/TX queue setup, sizing, sending and error debugging