Currently, torch and pytorch-lightning are specified as fixed dependencies, but often users installing this package already have these libraries (with specific versions) in their environments.
To improve flexibility and avoid unnecessary version conflicts:
- Update the requirements to use minimal version specifiers instead of fixed versions, e.g.
torch>=2, and likewise for pytorch-lightning if appropriate.
- Do not remove the dependencies entirely; they should remain with updated constraints.
- Add instructions in the documentation noting that users may need to install versions of
torch and/or pytorch-lightning that match their environment needs, and that the package only enforces a minimal version.
- Test that the minimal required versions are compatible with perspic, and note any important version constraints in the docs.
This will make perspic easier to integrate into projects and reduce potential versioning issues.
Currently,
torchandpytorch-lightningare specified as fixed dependencies, but often users installing this package already have these libraries (with specific versions) in their environments.To improve flexibility and avoid unnecessary version conflicts:
torch>=2, and likewise forpytorch-lightningif appropriate.torchand/orpytorch-lightningthat match their environment needs, and that the package only enforces a minimal version.This will make perspic easier to integrate into projects and reduce potential versioning issues.