Introducing Fleet: Cloud GPUs Without the Cloud Ops

by The Fleet Teamcompanygpulaunch

Machine learning engineers spend a startling amount of time not doing machine learning. Writing Dockerfiles. Debugging CUDA driver mismatches. Wrangling Kubernetes manifests to schedule a single training run. Waiting on a cluster that's idle 90% of the time but billed 100% of it.

Fleet exists to delete that work.

What Fleet is

Fleet is a cloud GPU platform for machine learning. You write ordinary Python on your own machine, and Fleet runs it on cloud GPUs — handling models, datasets, training jobs, and inference deployments for you. There's no container to build, no cluster to configure, and no infrastructure to babysit.

A full training run looks like this:

import fleet

client = fleet.Fleet()

model = client.models.from_huggingface("meta-llama/Llama-3.2-1B")
dataset = client.datasets.upload("./data.jsonl")

job = client.jobs.create(model=model, dataset=dataset)
job.monitor()              # stream live logs

fine_tuned = job.model()   # a new, deployable model

That's the whole thing. No Dockerfile, no YAML, no kubectl.

Why it's different

Three ideas shape everything we build:

Local-first. The best developer experience is the one you already have — your editor, your Python, your workflow. Fleet just adds the GPUs.
No ops tax. Nobody should have to learn Kubernetes to fine-tune a model. We hide the containers, schedulers, and drivers behind plain Python objects.
Pay for what you use. Billing is per-second. There are no seats, no subscriptions, and no idle costs — deployments even scale to zero when nothing is being served.

Who it's for

Fleet is built for ML engineers and researchers who want to train, fine-tune, and deploy models without becoming part-time infrastructure engineers. If you've ever shelved an idea because standing up the GPU environment wasn't worth it, Fleet is for you.

Getting started

You can be running on a cloud GPU in a few minutes:

pip install fleethq

From there, the getting-started guide walks you through your first training job and deployment. We can't wait to see what you build.

← ALL POSTS