Use RAG LLM Trick to Support Known Translations

[bshikin]

Using known_translations/{target_engine}.yaml with a collection of known correct translations.

Use chromadb as a vector DB for embeddings of known translations. Lookup closest translations and add them into the translation context on each query.

Example configuration:

known_translations:
  target_engine: godot
  embedding_function: krlvi/sentence-t5-base-nlpl-code_search_net
  result_count: 5
  user_request: |
    Remember this correct translation from Unity: 

${SOURCE}
```
is translated as:
```
${TARGET}
```

assistant_response: Certainly! I will remember this translation.

Where example content of the known translations file is:

translations:

• title: Ternary Operators source: | variable = {condition} ? (true_statement) : (false_statement); target: | variable = {translated_true_statement} if {translated_condition} else "{translated_false_statement}

  ---------------------------------------------------------------------------------------------------------------------

• title: Start() Function source: | private void Start() { {function_body} }

  target: | func _ready() -> void: { {translated_function_body} }

  ---------------------------------------------------------------------------------------------------------------------