Visibility for your Efficient Training

At MosaicML, we make fast, high-quality ML training accessible to everyone. Our MosaicML Cloud leverages algorithmic efficiency and effortless scale to bring the power of large deep learning models within reach of any enterprise. A key part of that accessibility is providing easy-to-use integrations with experiment tracking and training management solutions. Comet offers best-of-breed solutions to help you manage, visualize, and optimize your ML pipeline through dataset management, experiment tracking, a model registry, and production performance monitoring. Train your best model, using your organization’s data, on MosaicML Cloud, and keep track of the full pipeline with Comet.

Together, Comet and MosaicML enable faster, more efficient, and better instrumented model training. Through this combination of capabilities working in concert, your organization can deliver the best models for your business needs.

Start logging runs with Comet experiment tracking

With just a few lines of code and configuration in your MosaicML Cloud training runs, you can use Comet to log:

• Model-relevant metrics like loss
• System performance metrics like GPU utilization
• Hyperparameters
• Dependencies
• Debugging samples

and more. MosaicML Cloud’s “integrations” feature helps assemble your environment without all the boilerplate.

NOTE: Before getting started, sign up for MosaicML access and make sure you have completed setup for both MosaicML and Comet.

1. Add your Comet API key

Generate a Comet API Key at this link, then create an environment variable secret using MosaicML’s CLI:

mcli create secret env COMET_API_KEY=[YOUR COMET API KEY]

This command creates a secret which will mount your API key in the run execution environment as the variable COMET_API_KEY.

2. Add the Comet integration to your run configuration

In your run configuration YAML file, specify the workspace and project name within Comet where your run metrics should be logged:

integrations:
  - integration_type: comet_ml
    project_name: my_project
    workspace: my_workspace

3. Pass a CometLogger to your Composer Trainer

In your code, instantiate a CometMLLogger object, and pass it to your Composer Trainer:

from composer import Trainer
from composer.loggers import CometMLLogger

cometml_logger = CometMLLogger()

trainer = Trainer(
    model=model,
    train_dataloader=train_dataloader,
    eval_dataloader=eval_dataloader,
    loggers=[cometml_logger],
)

NOTE: The above is a Composer example, but the CometML integration will set up the environment for any code that uses the `comet_ml` python package.

4. You’re logging to Comet!

Log into your Comet dashboard and watch your training run data appear:

Try out MosaicML Cloud and Comet for yourself

(here's that MosaicML Cloud sign-up link again)

Ready to get started? Check out these helpful resources:

1. MosaicML Cloud
2. Comet docs
3. MosaicML docs
4. MosaicML Cloud + Comet integration page

If you have any questions, our teams are available on Comet Slack and MosaicML Slack communities.