machinewrapped/llm-subtrans v1.6.0

Support for terminology maps

on GitHub

Added support for a terminology map to be passed to the translator to ensure consistent translations, with an opt-in system that allows the LLM to add new terminology to the map itself, which will be fed-forward into future batches.

Supports editing the terminology map (on the fly, to some extent).

The UI has been restructured to give more space to the project settings panel, as it was getting a bit crowded.

The generated terminology map depends a lot on the model - some are parsimonious, some want to record every detail.

The default instructions contain a terminology_instructions section with the prompt that is appended to the request when automatic terminology mapping is enabled - feel free to tweak them until you get the level of detail you want.

The default fallback used by for other instructions files is currently identical, but custom terminology_instructions can be added to any instructions.

What's Changed

• Persist project progress on failures and standardize CLI translation … by @machinewrapped in #399
• Implement parent() method to fix NotImplementedError in SubtitleListModel by @PhilBug in #402
• Handle Ollama reasoning field in CustomClient; fall back to it when content is empty by @Copilot in #406
• Improve local LLM resilience by @machinewrapped in #407
• Fix init_project destructively re-batching when resuming an existing project by @machinewrapped in #411
• Add CLI progress logging and token usage reporting by @machinewrapped in #412
• Add opt-in terminology map for consistent translations by @machinewrapped in #413

New Contributors

• @PhilBug made their first contribution in #402

Full Changelog: v1.5.8...v1.6.0