modal-master

Expert Modal.com serverless cloud platform system with comprehensive Modal 1.0 SDK (May 2025) features, GPU functions (T4/L4/A10G/L40S/A100/H100/H200/B200), autoscaler configuration, @modal.concurrent/@modal.batched decorators, Sandboxes for isolated code execution, CloudBucketMount for S3/GCS, and production deployment patterns. PROACTIVELY activate for: (1) ANY Modal.com task, (2) GPU configuration with fallbacks and multi-GPU, (3) Autoscaler settings (min/max/buffer containers, scaledown_window), (4) Web endpoints (FastAPI, ASGI, WSGI, custom servers), (5) @modal.concurrent for request concurrency, (6) @modal.batched for dynamic batching, (7) Sandboxes for untrusted code execution, (8) Scheduling (Cron with timezone, Period), (9) Storage (Volumes with commit(), Dict with TTL, Queue, CloudBucketMount), (10) Parallel processing (.map(), .starmap(), .spawn(), .for_each()), (11) Container lifecycle (@modal.enter, @modal.method, @modal.exit), (12) Image building (uv_pip_install, run_function for model downloads), (13) Secrets and environment management, (14) Deployment and CI/CD with GitHub Actions, (15) Cost optimization and 2025 pricing. Provides: Modal 1.0 stable API patterns, GPU selection guide with per-second pricing, autoscaler tuning strategies, concurrency and batching for ML inference, Sandbox security patterns, CloudBucketMount for external data, complete CLI reference, debugging workflows, and production-ready configurations.