Generate Verl async DAPO training configurations with Megatron and FSDP2 options.
npx skills add https://github.com/Ascend/agent-skills --skill verl-async-dapo
Please help me install this Agent Skill. Skill: verl-async-dapo Source: https://github.com/Ascend/agent-skills/tree/main/skills/verl-async-dapo Command: npx skills add https://github.com/Ascend/agent-skills --skill verl-async-dapo
Verl async DAPO training configuration generation simplifies setup and eliminates repetitive scripting work.
Run the Verl config generator to produce a ready-to-run training script for your DAPO setup.
High-intent search queries and answers about installing and using this skill.
To generate Verl async DAPO training configs, run the config generator to produce ready-to-run Hydra-compatible YAML files, shell scripts, or command-line arguments for your setup.
Yes, Verl async DAPO configuration supports both Megatron and FSDP2 framework options, allowing you to specify parallel settings and trainer configurations across either framework.
Yes, Verl async DAPO supports preflight checks to validate your settings before execution, covering data, algorithm, parallel, rollout, and trainer configurations for single-shot DAPO runs.
SwanLab integration is configured through feature flag management during the Verl async DAPO setup, allowing you to enable tracking within the generated training scripts or YAML configs.
Automating DAPO training script setup is best handled by generating ready-to-run shell scripts with the Verl config generator, eliminating repetitive scripting work for reproducible experiments.
EnkeZMigrate RL training recipes from verl to Relax frameworks.
KappTech88Track ML experiments and visualize metrics across runs with SwanLab.
nishide-devCompose hierarchical Hydra configs with command-line overrides for PyTorch Lightning training runs.
ovachieverApply GRPO/RL training with TRL to optimize language models.
