Invalid PyTorch version pin and missing Sense2Vec model break fresh setup

📄 Issue Description
Summary

While setting up the project from a fresh clone, I ran into two issues that prevent the backend from running successfully out of the box.

1. Invalid PyTorch version in requirements.txt

The project pins PyTorch to a version that is not currently available on PyPI (e.g. torch==2.5.1).

This causes pip install -r requirements.txt to fail with:

ERROR: No matching distribution found for torch==2.x.x

On macOS (and generally on PyPI), the latest stable and available version appears to be torch==2.2.2, which works correctly with the rest of the stack.

2. Backend crash due to missing Sense2Vec model

On startup, the backend tries to load a local Sense2Vec model:

Sense2Vec().from_disk("s2v_old")

However, the s2v_old/ directory is not present in the repository, nor referenced in the setup documentation.
This results in a runtime crash on first run:

ValueError: Can't read file: s2v_old/strings.json

Impact

Fresh installs fail during dependency installation

Backend crashes even after successful installs

New contributors cannot run the project without manual debugging

Suggested approach (happy to help)

If this makes sense to the maintainers, I’d be happy to open a PR that:

Pins PyTorch to a valid, stable version available on PyPI

Handles a missing Sense2Vec model gracefully (e.g. optional loading or a clearer warning)

Optionally improves setup documentation around required models

Please let me know if you’d like me to proceed with a PR or if there’s a preferred approach.

Thanks!