Modal Integration

Note: Modal integration is currently in development. This page describes the planned functionality. RunPod is the recommended compute provider today.

Modal provides serverless GPU functions that run on-demand and bill per second. When available, Modal will complement RunPod by handling short-lived, bursty workloads.

Planned Features

Serverless GPU Functions

Instead of managing pods, deploy Python functions that run on GPU infrastructure:

# hubify_functions.py
import modal

app = modal.App("hubify-lab")

@app.function(gpu="H100", timeout=600)
def run_analysis(experiment_config: dict):
    """Runs on an H100 GPU, billed per second of execution."""
    import cobaya
    # Your analysis code here
    return results

When to Use Modal vs RunPod

Use Case	Recommended	Why
Long MCMC chains (hours)	RunPod	Persistent pod is cheaper for long runs
Quick analysis (< 10 min)	Modal	Per-second billing, no pod overhead
Figure generation	Modal	Short task, instant cold start
Batch inferences	Modal	Auto-scales across multiple GPUs
Interactive debugging	RunPod	SSH access, persistent environment
Training runs	RunPod	Stable long-running environment

Auto-Routing

When Modal is available, the orchestrator will automatically route experiments to the most cost-effective provider:

Short task (< 10 min) → Modal (per-second billing)
Long task (> 10 min) → RunPod (per-minute billing, cheaper at scale)
Bursty parallel tasks → Modal (auto-scaling)

Planned Setup

# Install Modal
pip install modal

# Authenticate
modal token new

# Connect to Hubify
hubify pod provider add modal --token $MODAL_TOKEN

# Deploy functions
hubify deploy functions --provider modal --file hubify_functions.py

Planned Configuration

# Set Modal as the default for short tasks
hubify pod config --short-task-provider modal

# Set the routing threshold
hubify pod config --modal-threshold 10m  # Tasks under 10 min → Modal

# View provider routing
hubify pod config --show-routing

Current Status

Modal integration is planned for Q3 2026. In the meantime:

• Use RunPod for all GPU compute
• The transition will be seamless when Modal support ships
• Existing experiments will continue to work on RunPod
• The orchestrator will automatically start routing to Modal when configured

Sign Up for Updates

To be notified when Modal integration is available:

hubify notifications subscribe --feature modal-integration