Deploys Verl training services on NPU clusters with RocketMQ, SeaWeedFS, and OpenResty.
npx skills add https://github.com/Ascend/agent-skills --skill verl-deploy
Please help me install this Agent Skill. Skill: verl-deploy Source: https://github.com/Ascend/agent-skills/tree/main/skills/verl-feature-deploy Command: npx skills add https://github.com/Ascend/agent-skills --skill verl-deploy
Verl Deploy enables one-click bootstrapping of Verl distributed training services on NPU clusters, automating environment validation, container provisioning, and monitoring configuration.
Generate the deployment scripts with the provided generator and deploy Verl inside the container by running start_verl.sh.
High-intent search queries and answers about installing and using this skill.
Deploy Verl on an NPU cluster using a one-click process that automates environment validation, Ray cluster setup, and SwanLab monitoring configuration. You generate deployment scripts with the provided generator and run start_verl.sh inside the container to launch training.
Yes, you can use a 7-bit feature mask to enable or disable specific acceleration optimizations for Megatron-based models like Qwen3-8B. This flexible masking allows fine-tuned control over distributed training performance on your provisioned containers.
The best way to configure Ray and SwanLab is through automated one-click deployment scripts that handle container orchestration and monitoring integration. The generated workflows coordinate pre-checks, docker operations, and SwanLab setup automatically.
Yes, you need environment checks before launching Verl. The deployment process enforces environment validation and pre-checks using provided scripts to ensure your NPU cluster is correctly provisioned and ready for container orchestration.
Yes, Verl deployment supports container orchestration through generated scripts and container-ready workflows. It coordinates docker operations, environment pre-checks, and SwanLab integration to ensure end-to-end deployment of distributed training on NPU clusters.
Deployment is specifically scoped to launching Verl on NPU clusters and configuring Megatron-based models. The provided scripts handle environment checks and container orchestration, but the 7-bit feature mask limits acceleration optimizations to supported model architectures like Qwen3-8B.
renneManage Proxmox VE clusters, SDN zones, and distributed firewall rules.
RheasilviaOrchestrate Megatron-LM training and SGLang rollouts for large-scale RL post-training.
qcmuuOrchestrate distributed ML training on multi-node Ray clusters.
ragnarokhaaTrain PyTorch models across GPUs and TPUs with a unified distributed API.
