product · May 18, 2026

Anyscale Introduces Post Training Skill for LLM Workloads

Anyscale released the /anyscale-workload-llm-post-training Agent Skill on May 14, 2026. The skill guides users through selecting post-training methods such as SFT, CPT, DPO, KTO, ORPO, SimPO, PPO, GRPO, or DAPO based on model, dataset, objective, and hardware. It generates configs for frameworks including LLaMA-Factory, SkyRL, and Ray Train while planning GPU memory, estimating run times like 36 hours on 4xL4 GPUs, and resolving dependencies. The output includes scripts, YAML files, requirements.txt, and READMEs compatible with Anyscale Jobs. Anyscale will exhibit at Microsoft Build Booth G201 on June 2-3, 2026.