Skip to content
What Tensorant is

Get started

What Tensorant is

What Tensorant does, how the console is laid out, where your data and GPUs live, and what you pay for.

Tensorant takes you from the data you already have to a fine-tuned model your application calls. You prepare and review examples in the console. Training and serving run on GPUs in your own RunPod account, and your data stays on your own RunPod network volume.

What it does

  1. Import your data. Upload files (TXT, Markdown, CSV, TSV, JSON, JSONL, YAML, HTML, PDF, DOCX, XLSX), paste text, or import a Hugging Face dataset. See Sources.
  2. Turn it into examples. Files that already hold questions and answers become examples directly. For documents, a recipe asks a model of your choice to write examples from the text.
  3. Review. Approve, edit or reject each example. Quality checks flag things like email addresses, phone numbers and very short answers. See Review.
  4. Freeze a version. Approved examples are frozen and split into training, validation and test sets. See Versions.
  5. Train and compare. The base model is scored on the test set first. Then a LoRA or QLoRA adapter is trained on one GPU, and the tuned model is scored the same way so you can compare the two. An experiment does all of this for you.
  6. Use it. Try the model in the Playground, publish it and call it through the OpenAI-compatible API, download the adapter, or publish it to Hugging Face.

The console at a glance

The sidebar follows the same path for the project you have open:

Group Pages
Home Progress, the next step, recent activity and anything holding a paid GPU
Data Sources, Recipes, Review, Versions
Tune Experiments, Runs, Evaluations
Use Playground, Deployments, API

The project switcher at the top of the sidebar opens or creates projects. At the foot, Settings shows your name, organization and role. It opens the settings of the project, the organization and your account, such as Members, Plan and billing, Connections, the theme and Sign out. The Activity button at the top of every page shows background work such as imports and evaluations.

What runs where

Where What is there
Tensorant The console and the API. It keeps your account, organizations, project settings, review states and scores, with references to the content on your volume.
Your RunPod account Every GPU: training runs, deployments and published models
Your RunPod network volume Your sources, examples, images, versions, evaluation answers and adapters
Your model endpoints The text you send them to generate examples or judge answers
Hugging Face Base models are downloaded from it; you can import datasets from it and publish adapters to it

Your content passes through Tensorant while you import, review or score it, and requests to a published model pass through Tensorant on their way to your GPU.

What it costs

  • GPUs. RunPod bills your account directly, at RunPod's prices. Every run and deployment has an hour limit and an hourly price limit, and nothing paid starts until you confirm it. An estimate is not a cap: it leaves out storage, CPU and disk fees and the time it takes to release a GPU. Once RunPod's billing has a run or deployment, Tensorant shows what RunPod actually billed for it. RunPod's billing can take a while to update, so you see the estimate until then.
  • Published models. An always-on model keeps a GPU running the whole time. A wake-on-request model runs on RunPod Serverless, which bills only while it runs but costs more per hour. See Publishing models.
  • Storage. RunPod bills your network volume for as long as it exists. Tensorant never deletes it. Its cost is shown on its own in Connections, not split across runs.
  • Model endpoints. Each endpoint's provider bills you for what you send it.
  • Your Tensorant plan. Free costs nothing, Starter is $99 a month and Pro is $499 a month. Enterprise has custom pricing. The plan sets how many imports, training runs, published models, API requests, projects, members and keys you get. It never includes GPU time. See Plans and billing.

Supported models

Tensorant trains and serves public, ungated chat models from Hugging Face in the Qwen, Llama, Mistral, gpt-oss, Phi-3 and GPT-NeoX families, and vision models such as Gemma 4 and Qwen3.8 that read images as well as text. Other models are allowed with a warning that they may not train or serve. See Supported models for the exact list and Vision models for image data.

Next steps