# Smaller Python ML containers

> Slim bases, same-layer cleanup, CPU-only PyTorch, pinned wheels, and the unglamorous work of keeping deploys fast.

- Category: Engineering
- Published: 2025-08-25
- Authors: Lalit Gupta
- Canonical: https://videodb.io/blog/python-ml-container-size
- HTML: https://videodb.io/blog/python-ml-container-size · Markdown: https://videodb.io/blog/python-ml-container-size.md
- Tags: docker, ml, buildkit

---
## What we saw
Python ML containers can become huge without anyone making an explicit decision. The common PyTorch install pulls CUDA libraries even for CPU-only workloads, which slows installs, bloats images, and can make serverless or autoscaled deploys slower.

## Why the images were so big
The default package path optimizes for broad hardware support. If your workload does not need a GPU, the CUDA baggage is pure deployment weight.

```dockerfile
# Before
pip install torch torchvision
# Can pull roughly 2.7 GB of CUDA-related packages

# After
pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
# CPU-only path, roughly 200 MB in the source note
```

## What we changed
Install CPU-only PyTorch before the rest of your requirements, and pin versions so a future upstream change does not alter your image unexpectedly.

```dockerfile
RUN pip install --no-cache-dir torch==2.6.0+cpu \
    --extra-index-url https://download.pytorch.org/whl/cpu && \
    pip install --no-cache-dir -r requirements.txt
```

Combine that with normal Docker hygiene: slim base images, fewer layers, same-layer cleanup for apt caches, `--no-install-recommends`, and `pip --no-cache-dir`.

## What it did to the numbers
| Install path | Approx package weight | Deploy effect |
|---|---:|---|
| Default PyTorch | ~2.7 GB CUDA libraries | Large image, slower install |
| CPU-only PyTorch | ~200 MB | Smaller image, faster deploy |

## What we would tell another team
Container size is often dependency selection disguised as infrastructure. If a workload is CPU-only, make that choice explicit in the Dockerfile.

## Cite this work

Lalit Gupta, "Smaller Python ML containers", VideoDB Labs, August 2025.

```bibtex
@article{gupta2025smaller,
  author = {Lalit Gupta},
  title = {Smaller Python ML containers},
  journal = {VideoDB Labs},
  year = {2025},
  month = {aug},
  note = {https://videodb.io/blog/python-ml-container-size},
}
```