modal-gpu

Run Python code on cloud GPUs using Modal serverless platform. Use when you need A100/T4/A10G GPU access for training ML models. Covers Modal app setup, GPU...

Security Scan

The instructions legitimately describe running Python on Modal GPUs, but the skill's metadata omits the credentials and environment access the docs require (Modal tokens and optional HF tokens), creating an incoherence you should understand before installing.

Purpose & Capability

The skill's name and documentation describe Modal GPU usage (pip install modal, modal token set, using Modal functions and volumes). That capability aligns with the stated purpose. However, the SKILL metadata declares no required credentials or primary credential even though the runtime instructions explicitly instruct the user to set a Modal token (and describe using HuggingFace tokens/secrets for private data). This mismatch between declared requirements and runtime instructions is a design omission.

Instruction Scope

SKILL.md stays on-topic: it outlines creating Modal apps, building images, downloading data inside remote functions, and returning results. It does instruct running `modal token set --token-id <id> --token-secret <secret>` and references HF_TOKEN usage and Modal secrets. The instructions do not ask to read unrelated files or exfiltrate data, but they rely on secrets/credentials that are not declared in the metadata.

Install Mechanism

This is instruction-only with no install spec and no code files. The only installation guidance is `pip install modal`, which is appropriate and low risk for this purpose.

Credentials

The runtime docs require setting a Modal token (and show patterns for using a HuggingFace token for private datasets), yet the skill declares no required environment variables or primary credential. That omission hides that tokens/credentials are necessary and materially affect trust. Requesting tokens for the platform and optional dataset access is reasonable for the skill's function, but the metadata should declare them. Users should be aware they will need to provide secrets and should limit permissions.

Persistence & Privilege

always is false and there is no install writing persistent configuration. The skill does mention creating Modal secrets and volumes, which are normal Modal features. There is no indication the skill requests elevated platform privileges beyond normal Modal usage.

What to consider before installing

This skill appears to be what it claims (how to run training on Modal GPUs) but the metadata doesn't list the credentials the instructions use. Before installing or running: 1) Expect to provide a Modal token (and optionally a HuggingFace token for private datasets); only provide tokens you trust and scope them with least privilege. 2) Prefer creating Modal secrets (Modal's secret store) rather than pasting tokens into scripts. 3) Verify the skill's origin and review any code you run on Modal (especially user-provided scripts) for data exfiltration or unexpected network calls. 4) Test with non-sensitive/limited-permission tokens first and rotate tokens after testing. 5) If you need absolute assurance, ask the publisher to update the skill metadata to declare required credentials (e.g., PRIMARY_ENV or requires.env) so the requirements are explicit.

Modal GPU Training

Overview

Modal is a serverless platform for running Python code on cloud GPUs. It provides:

• Serverless GPUs: On-demand access to T4, A10G, A100 GPUs
• Container Images: Define dependencies declaratively with pip
• Remote Execution: Run functions on cloud infrastructure
• Result Handling: Return Python objects from remote functions

Two patterns:

• Single Function: Simple script with @app.function decorator
• Multi-Function: Complex workflows with multiple remote calls

Quick Reference

Topic	Reference
Basic Structure	Getting Started
GPU Options	GPU Selection
Data Handling	Data Download
Results & Outputs	Results
Troubleshooting	Common Issues

Installation

pip install modal
modal token set --token-id <id> --token-secret <secret>

Minimal Example

import modal

app = modal.App("my-training-app")

image = modal.Image.debian_slim(python_version="3.11").pip_install(
    "torch",
    "einops",
    "numpy",
)

@app.function(gpu="A100", image=image, timeout=3600)
def train():
    import torch
    device = torch.device("cuda")
    print(f"Using GPU: {torch.cuda.get_device_name(0)}")

    # Training code here
    return {"loss": 0.5}

@app.local_entrypoint()
def main():
    results = train.remote()
    print(results)

Common Imports

import modal
from modal import Image, App

# Inside remote function
import torch
import torch.nn as nn
from huggingface_hub import hf_hub_download

When to Use What

Scenario	Approach
Quick GPU experiments	gpu="T4" (16GB, cheapest)
Medium training jobs	gpu="A10G" (24GB)
Large-scale training	gpu="A100" (40/80GB, fastest)
Long-running jobs	Set timeout=3600 or higher
Data from HuggingFace	Download inside function with hf_hub_download
Return metrics	Return dict from function

Running

# Run script
modal run train_modal.py

# Run in background
modal run --detach train_modal.py

External Resources

• Modal Documentation: https://modal.com/docs
• Modal Examples: https://github.com/modal-labs/modal-examples