Implement offline dataset processing and local cache management for desktop users

[SahilKumar75]

Summary

Since TuneOS is running as a local desktop app, we should provide tools to manage the Hugging Face local cache and process datasets entirely offline without requiring constant internet access.

Scope

• What changed: Settings/Cache UI, Dataset processing backend
• What did not change: Core training loop
• Related issue: N/A

Product Impact

Users on strict corporate networks or airplanes can prepare and tokenize their data without being blocked by network connectivity checks from Hugging Face Hub.

Technical Notes

Investigate HF_DATASETS_OFFLINE and TRANSFORMERS_OFFLINE environment variables.

Validation

• Application starts and trains completely offline with cached models
• Cache management UI displays downloaded models