Modal and Runpod
Optional platforms for GPU, burst workers, or pod-style execution. This repo does not ship proprietary orchestration; integration is your Modal app or Runpod template calling the same containers / CLI documented in Docker.
Modal
Fit: Strong for async translation workers (queue + modal.Function / .map), custom images, and volume mounts for caches.
Pattern:
- Build or reference a base image derived from
runtime-cpu/runtime-vision(see Docker image profiles and the repoDockerfile). - Run
doctranslate translatewith input/output on Modal volumes or download/upload from object storage. - Keep secrets (API keys) in Modal secrets, not in images.
Why not the primary OSS doc target: Modal uses Python-first APIs; many users want plain Docker + YAML first—see Deploy on Cloud Run.
Runpod
Fit: Possible — useful when you want GPU (e.g. ONNX GPU extras in pyproject.toml cuda / directml) or per-job pods with large disks.
Pattern:
- Container image:
runtime-visionfor full optional stack (OCR, Hyperscan glossary path) or customruntime-cpuwithINCLUDE_OCR=1(seepyproject.tomloptional extras). - Mount host or network volume for
~/.cache/doctranslatewhere supported. - Drive jobs via Runpod serverless handler that shells
doctranslate translateor importsdoctranslate.api.