Skip to content

v1.7.0a1

Choose a tag to compare

View all tags
@zh-plus zh-plus released this 25 May 03:37
· 13 commits to master since this release
1.7.0
45be59c

Add a token-efficient lean translation pipeline, broaden model-provider routing, and refresh the configuration surface for mixed-model workflows.

New Features:

  • Add LeanTranslator with compact anchored prompts, sliding context, checkpoint resume, retry handling, binary-split recovery, and atomic fallback for missing lines.
  • Add lean mode integration to LRCer through TranslationConfig(translate_mode="lean"), including optional Context Review and separate CR/translation chatbots.
  • Add LiteLLM support as an optional provider gateway via openlrc[litellm] and litellm:<provider/model> routing.
  • Add ModelConfig.extra_body passthrough for provider-specific request parameters across OpenAI-compatible, Anthropic, Gemini, and LiteLLM backends.

Other Changes:

  • Replace legacy string-based translation config fields with ModelConfig objects for primary, retry, and Context Review chatbots.
  • Decouple model capability limits from the static model registry; max_tokens and context_window are now opt-in runtime caps.
  • Improve custom-provider and provider-prefix inference for OpenAI-compatible, Anthropic, Gemini, LiteLLM, and third-party model names.
  • Restrict lingua language detection to the supported language set and add anchor-based validation for lean translation output.
  • Split media helpers from utils into media_utils to keep heavyweight media dependencies isolated.
  • Centralize live API test configuration, add LeanTranslator and LiteLLM regression coverage, and update CI to exercise the new optional litellm extra.
  • Refresh README examples for lean mode, mixed-model translation, live test configuration, and the current local-LLM/provider support status.
Assets 2
Loading