// MODELS

Your models,one object away.

Bring a base model from HuggingFace or your own weights, fine-tune it, and ship the result. Fleet stores, versions, and dedupes every model so you never upload the same bytes twice.

GET STARTED →READ THE DOCS

FLEET // PYTHON
import fleet
 
client = fleet.Fleet()
 
model = client.models.from_huggingface(
    "meta-llama/Llama-3.2-1B"
)
 
local = client.models.upload(
    "./my-model/"
)  # content-hashed

// WHAT YOU GET

ANY SOURCE

Pull a base model straight from a HuggingFace repo ID, or upload your own weights from a local directory.

CONTENT-HASHED

Fleet hashes weights before upload. Re-uploading the same model is instant and free, with no duplicate storage.

VERSIONED OUTPUTS

Every training run produces a new model you can deploy, download, or keep fine-tuning from.

DEPLOY IN ONE CALL

Turn any ready model into a live, scalable inference endpoint with a single method.

READY TO LAUNCH?

Join the engineers already training on Fleet.

GET STARTED FOR FREE →

★ FLEET PLATFORM ONLINE★ GPU SCHEDULING ACTIVE★ ALL SYSTEMS GO★ INFERENCE ENDPOINTS READY★ MISSION CONTROL STANDING BY★ LAUNCH IN T-MINUS ZERO★ FLEET PLATFORM ONLINE★ GPU SCHEDULING ACTIVE★ ALL SYSTEMS GO★ INFERENCE ENDPOINTS READY★ MISSION CONTROL STANDING BY★ LAUNCH IN T-MINUS ZERO★ FLEET PLATFORM ONLINE★ GPU SCHEDULING ACTIVE★ ALL SYSTEMS GO★ INFERENCE ENDPOINTS READY★ MISSION CONTROL STANDING BY★ LAUNCH IN T-MINUS ZERO★ FLEET PLATFORM ONLINE★ GPU SCHEDULING ACTIVE★ ALL SYSTEMS GO★ INFERENCE ENDPOINTS READY★ MISSION CONTROL STANDING BY★ LAUNCH IN T-MINUS ZERO★ FLEET PLATFORM ONLINE★ GPU SCHEDULING ACTIVE★ ALL SYSTEMS GO★ INFERENCE ENDPOINTS READY★ MISSION CONTROL STANDING BY★ LAUNCH IN T-MINUS ZERO★ FLEET PLATFORM ONLINE★ GPU SCHEDULING ACTIVE★ ALL SYSTEMS GO★ INFERENCE ENDPOINTS READY★ MISSION CONTROL STANDING BY★ LAUNCH IN T-MINUS ZERO