---
tags:
- docker
- environment
- codiffact
---

# CoDiffAct environment-only image

Private Linux/amd64 Docker archive. Contains installed runtime dependencies; no project, datasets, pretrained weights or credentials. Each dependency retains its own license; this card does not relicense them.

Download all parts, verify SHA256SUMS, then load:

```bash
sha256sum -c SHA256SUMS
cat codiffact-env_20261006-cu128.tar.gz.part-* | gzip -dc | docker load
docker run --rm --network none --gpus all codiffact-full:20261006-cu128 python /opt/codiffact/smoke_test.py --gpu
```

See model_setup.md for public model sources, paths and mounts. validation.json records offline dependency, OSMesa, audio/video and CUDA smoke checks, not full model inference/training. A compatible host NVIDIA driver and NVIDIA Container Toolkit are required.

The exact installed-prefix snapshot is embedded in the archive. requirements.snapshot.txt is an inventory, not a universal pip installation lock. Rebuilding with prepare_environment.py requires the recorded source environment; Dockerfile alone is not a complete from-PyPI recipe.