Replicate Version control for machine learning

Lightweight, open source

01 Never lose your work

Train as usual, and Replicate will automatically save your code and weights to Amazon S3 or Google Cloud Storage.

02 Go back in time

Get back the code and weights from any checkpoint if you need to re-train or get the model weights out.

03 Toward reproducibility

It is really hard to run and re-train ML models. We are working to fix that, but we need your help to make it a reality.

04 How it works

Just add two lines of code. You don’t need to change how you work.

Replicate is a Python library that uploads files and metadata (like hyperparameters) to Amazon S3 or Google Cloud Storage.

You can get the data back out using the command-line interface or a notebook.

import torch
import replicate
def train():
    # Save training code and hyperparameters
    experiment = replicate.init(path=".", params={...})
    model = Model()
    for epoch in range(num_epochs):
        # ...
        torch.save(model, "model.pth")
        # Save model weights and metrics
        experiment.checkpoint(path="model.pth", metrics={...})

import torch
import replicate
def train():
    # Save training code and params
    experiment = replicate.init(
        path=".",
        params={...},
    )
    model = Model()
    for epoch in range(num_epochs):
        # ...
        torch.save(model, "model.pth")
        # Save model weights and metrics
        experiment.checkpoint(
            path="model.pth",
            metrics={...},
        )

05 Open source & community-built

We’re trying to pull together the ML community so we can build this foundational piece of technology together. Learn more.

06 It's just plain old files on S3

All the data is stored on your own Amazon S3 or Google Cloud Storage as tarballs and JSON files. There’s no server to run. Learn more.

07 Works with everything

Tensorflow, PyTorch, scikit-learn, XGBoost, you name it. It’s just saving files and dictionaries – export however you want.

Throw away your spreadsheet

Your experiments are all in one place, with filter and sort. Because the data’s stored on S3, you can even see experiments that were run on other machines.

$ replicate ls --filter "val_loss<0.2"
EXPERIMENT   HOST         STATUS    BEST CHECKPOINT
e510303      10.52.2.23   stopped   49668cb (val_loss=0.1484)
9e97e07      10.52.7.11   running   41f0c60 (val_loss=0.1989)

Analyze in a notebook

Don’t like the CLI? No problem. You can get all your experiments out and do meta-analysis on your results from within a notebook. Learn more.

Compare experiments

It diffs everything, all the way down to versions of dependencies, just in case that latest Tensorflow version did something weird.

$ replicate diff 49668cb 41f0c60
Checkpoint:       49668cb     41f0c60
Experiment:       e510303     9e97e07
Params
learning_rate:    0.001       0.002
Python Packages
tensorflow:       2.3.0       2.3.1
Metrics
train_loss:       0.4626      0.8155
train_accuracy:   0.7909      0.7254
val_loss:         0.1484      0.1989
val_accuracy:     0.9607      0.9411

Get back your code and weights

Replicate lets you get back to any point you called experiment.checkpoint() so, you can re-train models and get your model weights out.

$ replicate checkout f81069d
Copying code and weights to working directory...
If you want to run this experiment again, this is how it was run:
  python train.py --learning_rate=0.2

Load models for inference

You can load a specific version of your model in Python to run inferences – either from within a notebook or from another program running inferences.

A platform to build upon

Replicate is intentionally lightweight and doesn’t try to do too much. Instead, we give you Python and command-line APIs so you can integrate it with your own tools and workflow.

Get started

or,

learn more about how Replicate works