Deploy to Production

Deploy your models to production environments with confidence using the right deployment strategies and configurations. This guide focuses on model deployment patterns, authentication modes, and configuration best practices.

Deployment Types

Ephemeral Deployments

For development and testing, use ephemeral deployments with fal run.

fal run MyApp::path/to/myapp.py

Once you kill the fal run process in your terminal, the ephemeral deployment will be destroyed.

Persistent Deployments

To permanently deploy your application or update/redeploy existing one, you can use the fal deploy command.

fal deploy --auth private

Machine Types

You can specify the machine type of your app using the machine_type parameter in your fal.App class.

For GPU machines, you can also specify the number of GPUs you want to use with the num_gpus option.

class MyApp(fal.App):
    machine_type = "GPU-A100"
    num_gpus = 1
    ...

Or you may specify the machine type in the fal deploy command.

fal deploy --machine-type GPU-A100 --num-gpus 1

Machine Type Options

Value	Description
XS	0.50 CPU cores, 512MB RAM
S	1 CPU core, 1GB RAM (default)
M	2 CPU cores, 2GB RAM
L	4 CPU cores, 15GB RAM
GPU-A6000	10 CPU cores, 18GB RAM, 1 GPU core (48GB VRAM)
GPU-A100	12 CPU cores, 60GB RAM, 1 GPU core (40GB VRAM)
GPU-H100	12 CPU cores, 112GB RAM, 1 GPU core (80GB VRAM)
GPU-H200	12 CPU cores, 112GB RAM, 1 GPU core (141GB VRAM)
GPU-B200	24 CPU cores, 112GB RAM, 1 GPU core (192GB VRAM)

Multiple Machine Types

Allow your app to use multiple machine types for a larger pool of available machines:

class MyApp(fal.App):
    machine_type = ["GPU-A100-40G", "GPU-A100-80G"]

Rollout Strategies

Your app could be deployed using one of two strategies:

• recreate: default, instantly switch the app to the new revision.
• rolling: doesn’t switch the app to the new one until there is at least 1 runner in the new revision.

You can specify the strategy using the --strategy flag, e.g.

fal deploy --strategy rolling

Authentication Modes

Your app could be deployed in one of three authentication modes:

• private: default, your app is visible only to you and/or your team.
• shared: everyone can see and use your app, the caller pays for usage. This is how all of the apps in our Model Gallery work.
• public: everyone can see and use your app, the app owner (you) is paying for it.

Use fal deploy’s --auth flag or fal.App’s app_auth to specify your app’s authentication mode, e.g.

class MyApp(fal.App):
    auth_mode = "shared"

fal deploy --auth shared

To change the mode just redeploy the app.

Best Practices

• Choose rolling strategy for production deployments to ensure zero downtime
• Use appropriate authentication modes based on your use case and cost considerations
• Test thoroughly with ephemeral deployments before permanent deployment
• Monitor your deployments using the fal dashboard and performance monitoring tools

Managing Deployed Models

For managing your deployed models (listing, deleting, monitoring), see Manage Deployments and Monitor Performance.