Library Usage

Simple text translation

from pypolyglot import translate

result = await translate("Hello world", "en", "ja")
print(result.translation)  # こんにちは世界

The translate function handles chunking, batching, and retry automatically. For long text, it splits into segments, translates each, and reassembles the result.

Markdown translation

from pypolyglot import translate_markdown

md = """## Features

Local GPU translation with **zero cloud dependency**.

```python
result = await translate("Hello", "en", "ja")

"""

result = await translate_markdown(md, “en”, “fr”) print(result.markdown)

Markdown translation preserves:
- Code blocks (fenced and inline)
- Tables
- HTML tags and attributes
- URLs, badges, and image references
- Heading structure

## Multi-language translation

Translate into 7 languages at once:

```python
from pypolyglot import translate_all

result = await translate_all(md, source_lang="en")
for r in result.results:
    print(f"{r.name}: {r.status} ({r.duration_ms:.0f}ms)")

Default languages: Japanese, Chinese (Simplified), Spanish, French, Hindi, Italian, Portuguese (BR).

Target specific languages:

result = await translate_all(md, target_langs=["ja", "es", "fr"])

Translation options

Custom model

from pypolyglot import translate, TranslateOptions

result = await translate("Hello", "en", "ja",
    TranslateOptions(model="translategemma:4b"))

Custom glossary

Override translations for specific terms:

from pypolyglot import translate, TranslateOptions, GlossaryEntry

result = await translate("Deploy the Widget", "en", "ja",
    TranslateOptions(glossary=[
        GlossaryEntry("Widget", {"ja": "ウィジェット"})
    ]))

polyglot-gpu includes a built-in software glossary with 12 technical terms (Architecture, Pipeline, Deploy, Library, Framework, Build, Release, Branch, Repository, Merge, Token, Adoption). These are injected automatically when the terms appear in the source text.

Streaming

result = await translate("Hello world", "en", "ja",
    TranslateOptions(on_token=lambda t: print(t, end="")))

Caching

Segment-level caching is enabled by default. Cache uses Levenshtein fuzzy matching — if a segment is similar enough to a previously translated one, the cached translation is reused.

from pypolyglot import translate_markdown, TranslateMarkdownOptions

# Disable caching
result = await translate_markdown(md, "en", "ja",
    TranslateMarkdownOptions(cache=False))

Cache files are stored as .polyglot-cache.json in the working directory.