Skip to content

User guide

The user guide explains each part of Tuneline Studio by topic: how pipelines, steps and artifacts fit together, how to build and run a pipeline, where it can run and what that costs, how to read the results, and how to put a trained model to use. If you are new, start with Getting started to run your first pipeline, then come back here for the details. For exact settings of every step, see the API reference.

Basics

  • Core concepts: pipelines, steps, artifact types, runs, caching, targets and keys, and how they relate.
  • The pipeline editor: the canvas, the step palette, links, the inspector, undo, autosave and validation.
  • Datasets: uploading files, using Hugging Face datasets or URLs, format detection, previews and versions.
  • Models: the catalog of open models, Hub search, licenses, gated models and memory needs.
  • Hardware and costs: the hardware advisor, which GPUs fit your model, and estimated time and cost per provider.

Running

  • Running pipelines: the Run dialog, choosing where to run, GPU choice, time and cost limits, and cached steps.
  • Your own computer: connecting a computer as a worker so My computer runs start by themselves.
  • Watching runs: live status on the canvas, the run page (logs, metrics, evaluation, artifacts), comparing runs, cancel and retry.

Results

  • Evaluation and quality gates: scoring models on a test set or benchmarks, base-vs-tuned comparison, and rules that stop a weak model.
  • Registry: versioned models with stages, scores, model cards and lineage.
  • Deployments: OpenAI-compatible endpoints on RunPod, Hugging Face or Modal, and free ways to use your model.
  • Playground: chatting with your endpoints or any Hub model, side-by-side comparison, and golden sets.

Setup and safety

  • API keys and security: which provider keys you need, how to create them, and how they are stored and protected.
  • Pipeline YAML: the YAML view of a pipeline, editing it, and running it from the command line.

Help

  • Troubleshooting: common problems, the messages you see, and how to fix them.
  • FAQ: short answers to frequent questions about cost, privacy, hardware and models.