Skill UI

This workflow guides users through the complete lifecycle of customizing large language models (LLMs). It covers uploading training data, initiating fine-tuning jobs, monitoring training progress, managing model versions, and finally deploying the optimized model to a dedicated, autoscaling inference endpoint using the Together AI API. Ideal for production-level customization.

SkillOpt-Sleep simulates an agent's nightly 'sleep' cycle to continuously improve performance without requiring model retraining. It harvests historical sessions, mines recurring tasks, and replays them offline. Learning and insights are then consolidated into the agent's memory and skills (CLAUDE.md/SKILL.md), proposing bounded, validated edits that the user must explicitly adopt, ensuring measurable, sustained improvement over time.

This system creates and registers high-quality, reusable synthetic voices (voice clones) from local audio samples. It operates under a strict, consent-first workflow, requiring explicit speaker consent, detailed use permissions, and source verification before cloning. It is designed for professional Text-to-Speech (TTS) applications, IVR systems, or training content development where a consistent, branded voice is essential.

This skill manages the entire lifecycle of long-running deep learning training and inference jobs deployed on remote, ephemeral computing environments (e.g., AutoDL, RunPod, bare SSH, Slurm, K8s). It provides critical capabilities like resilient checkpointing, managing spot instance preemption, debugging resource failures (OOM, NaN), and ensuring safe billing teardown, making it essential for professional ML Ops on rented resources.