Summary

• Adds LiteLLM as a new translation service provider, giving users access to 100+ LLM providers (Anthropic, Bedrock, Vertex AI, Cohere, Mistral, etc.) through a single unified interface.
• Follows the existing translator class pattern. Drop-in, no changes to existing providers.

Motivation

PDFMathTranslate currently supports individual providers (OpenAI, Gemini, Groq, Deepseek, etc.), each requiring a separate translator class. LiteLLM is a lightweight Python SDK that provides a unified completion() interface across 100+ providers. Users specify the provider via the model string (e.g. anthropic/claude-sonnet-4-5, bedrock/anthropic.claude-v2, vertex_ai/gemini-pro) and LiteLLM routes the call to the correct provider API. This eliminates the need to add individual translator classes for each new provider.

Changes

• pdf2zh/translator.py - new LiteLLMTranslator class extending BaseTranslator
• pdf2zh/converter.py - registered in import list and translator loop
• pdf2zh/gui.py - registered in import, service_map, babeldoc_translate_file loop, and isinstance check for rich text support
• pdf2zh/pdf2zh.py - registered in import and translator loop
• pyproject.toml - added litellm>=1.55.0,<1.85 as optional dependency (pip install pdf2zh[litellm])

Implementation details

• Optional dependency: litellm is an optional extra (like argostranslate), so the base install is unaffected. Lazy-imported with a clear error message if not installed.
• drop_params=True: Silently drops provider-unsupported kwargs (e.g. strict, seed), preventing cross-provider failures.
• Streaming: Always enabled, matching OpenAITranslator behavior.
• Retry on rate limits: Uses tenacity with exponential backoff on litellm.RateLimitError.
• Think-tag filtering: Strips <think> tags from reasoning model outputs (DeepSeek-R1, QwQ, etc.).
• Rich text support: Enabled in babeldoc mode via isinstance check alongside OpenAITranslator.
• Provider auth: LiteLLM reads provider-specific env vars automatically (ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, etc.).

Service table entry

Example usage

Use -s litellm:model to specify the provider and model:

pip install pdf2zh[litellm]

# Translate using Anthropic Claude
export ANTHROPIC_API_KEY="sk-..."
pdf2zh example.pdf -s litellm:anthropic/claude-sonnet-4-5

# Translate using AWS Bedrock
export AWS_ACCESS_KEY_ID="..." AWS_SECRET_ACCESS_KEY="..." AWS_REGION_NAME="us-east-1"
pdf2zh example.pdf -s litellm:bedrock/anthropic.claude-v2

# Translate using Google Vertex AI
export VERTEX_PROJECT="my-project" VERTEX_LOCATION="us-central1"
pdf2zh example.pdf -s litellm:vertex_ai/gemini-pro

Or specify the model with environment variables:

export LITELLM_MODEL=anthropic/claude-sonnet-4-5
export ANTHROPIC_API_KEY="sk-..."
pdf2zh example.pdf -s litellm

GUI: Select "LiteLLM" from the service dropdown, set LITELLM_MODEL to your desired model string (e.g. anthropic/claude-sonnet-4-5). Set the provider's API key as an environment variable before launching the GUI.

Tests

Live E2E test against Anthropic Claude (anthropic/claude-sonnet-4-5):

>>> t = LiteLLMTranslator('en', 'zh', 'anthropic/claude-sonnet-4-5')
>>> t.do_translate('Hello world')
'你好世界'

>>> t.do_translate('Machine learning is transforming the world.')
'机器学习正在改变世界。'

>>> t.do_translate('The equation {{v0}} shows that {{v1}} is proportional to {{v2}}.')
'方程 {{v0}} 表明 {{v1}} 与 {{v2}} 成正比。'

Formula placeholders ({{v0}}, {{v1}}, {{v2}}) are preserved correctly in translated output.

Risk / Compatibility

• Additive only. Existing providers untouched.
• litellm is an optional extra. Base install unaffected.
• Pinned to >=1.55.0,<1.85 for stability.