fal-serverless-guide

## Quick Reference

| Machine Type | GPU | VRAM | Use Case |
|--------------|-----|------|----------|
| `GPU-T4` | T4 | 16GB | Dev, small models |
| `GPU-A10G` | A10G | 24GB | 7B-13B models |
| `GPU-A100` | A100 | 40/80GB | 13B-70B models |
| `GPU-H100` | H100 | 80GB | Cutting-edge |

| App Attribute | Purpose | Example |
|---------------|---------|---------|
| `machine_type` | GPU selection | `"GPU-A100"` |
| `requirements` | Dependencies | `["torch", "transformers"]` |
| `keep_alive` | Warm duration | `300` (5 min) |
| `min_concurrency` | Min instances | `0` (scale to zero) |
| `max_concurrency` | Max parallel | `4` |

| Command | Purpose |
|---------|---------|
| `fal deploy app.py::MyApp` | Deploy to fal |
| `fal run app.py::MyApp` | Run locally |
| `fal logs <app-id>` | View logs |
| `fal secrets set KEY=value` | Set secrets |

## When to Use This Skill

Use for **custom model deployment**:
- Deploying custom ML models on fal infrastructure
- Configuring GPU instances and scaling
- Setting up persistent storage for model weights
- Creating multi-endpoint apps
- Managing secrets and environment variables

**Related skills:**
- For API integration: see `fal-api-reference`
- For optimization: see `fal-optimization`
- For using hosted models: see `fal-model-guide`

---

# fal.ai Serverless Deployment Guide

Complete guide to deploying custom ML models on fal.ai's serverless infrastructure.

## Overview

fal serverless provides:
- Automatic scaling from zero to thousands of instances
- GPU support (T4, A10G, A100, H100, H200, B200)
- Persistent storage for model weights
- Secrets management
- Real-time logs and monitoring
- Pay-per-use pricing

## Installation

```bash
pip install fal
```

## Authentication

```bash
# Login to fal
fal auth login

# Or set API key
export FAL_KEY="your-api-key"
```

## Basic App Structure

```python
import fal
from pydantic import BaseModel

class RequestModel(BaseModel):
    """Input schema for your endpoint"""
    prompt: str
    ma