Fine-tune and deploy open source models

Fine-tune, deploy, and invoke Stable Diffusion, FLAN-T5, and more – all from an easy-to-use API.

Get an API key Colab demo

main.py

Run on Colab

Everything you need to make models your own.

Dataset API

Programmatically upload, manage and validate datasets for tuning your models. Learn more

Fine-tuning API

Fine-tune open source models with ready-to-use training scripts. Just bring your data. Learn more

Deployment API

Deploy fine-tuned models in one click onto serverless GPUs, no infra required. Learn more

Serverless functions

Write serverless Python functions and integrate you model with any of your applications. Learn more

Fine-tuning API

Fine-tune any model

Just bring your data

Ready-to-use training scripts make it easy to get started and an extensible API lets you adjust parameters as you go. Learn more

Stable Diffusion training config

1234567

{
  "batch_size" : 4, 
  "learning_rate" : 1e-6, 
  "learning_rate_schedule" : cosine,
  "warmup_percentage" : 10, 
  "requires_grad" : ["unet", "vae"]
}

Programmatically fine-tune models

Embed fine-tuning into your workflow or application with Blueprint’s feature-rich API. Learn more

~shell

from baseten.training import FinetuningRun, LocalPath, DreamboothConfig

input_data = LocalPath('./home/my-dog')

config = DreamboothConfig( instance_prompt='photo of sks charlie', input_dataset=input_data, num_train_epochs=10

my_run = finetune.create("model_name", config)

🔮 Fine-tune run is in progess...$ my_run.stream_logs()

Uploading data100%21.0M/492M

Progress69%1250/1300 steps

Deployment API

Deploy and scale your models

Auto deploy fine-tuned models.

Quickly auto deploy fine-tuned models onto auto-scaling, serverless GPUs. No infra needed. Learn more

Model 3Deploy

Model 2Deploy

Avatar generator

Fine-tunedDeployed

StatusActive

Runtime42m

API calls142/s

Idle shutdown30m

Replicas2

import baseten import StableDiffusionPipeline

# Invoke your model
model = StableDiffusionPipeline(model_id="rwnod2q")
image url = model("a photo of sks dog playing in the snow")
image.save("dog-snow.png")

Only pay for what you use.

Blueprint’s serverless GPU infrastructure auto-scales based on traffic to your model. See pricing

Serverless functions

Ship your models

Build APIs using serverless routes. Build endpoints without spinning up new services. Learn more

Test your models in an interactive playground. Test and introspect your models. Learn more

app.baseten.co/blueprint/your_project

Get started

Building with Blueprint

I fine-tuned FLAN-T5. Can it cook?

I used Blueprint to train FLAN-T5 to generate a recipe for any dish I could dream up.

How it's built View project

Build an avatar generator

In less than 200 lines of code you'll build a simple version of an avatar generation app like Lensa using Blueprint.

View the code

Fine-tuning with Dreambooth in Figma

See how to build a FigJam plugin to fine-tune and invoke Stable Diffusion with Dreambooth directly from Figma.

How it's built

Built on Truss and Baseten

Blueprint is built on Truss' open-source standard for packaging models and on Baseten's scalable model serving infrastructure, trusted by companies including Patreon, Pipe, Motive, and Laurel.

Ready to get started?

Start your project with 4 hours of free fine-tuning and model serving GPU credits.

Start building Explore the docs