TGStat
TGStat
Введите текст для поиска
Расширенный поиск каналов
  • flag Russian
    Язык сайта
    flag Russian flag English flag Uzbek
  • Вход на сайт
  • Каталог
    Каталог каналов и чатов Поиск каналов
    Добавить канал/чат
  • Рейтинги
    Рейтинг каналов Рейтинг чатов Рейтинг публикаций
    Рейтинги брендов и персон
  • Аналитика
  • Поиск по публикациям
  • Мониторинг Telegram
mohaka channel

4 Nov, 20:41

Открыть в Telegram Поделиться Пожаловаться

Leaving an OpenAI-compatible Model field blank discovers the first server-advertised model. Gemini continues to use its default model when blank.
Gemini errors identify invalid keys, unavailable models, and detected region/IP restrictions with actionable explanations.
Arabic, Gemini and language-cache fixes
Arabic translation prompts now enforce Modern Standard Arabic, and Gemini reports incomplete token-limited responses and regional/network failures clearly. The default Gemini model is gemini-3.8-flash.
Translation results are cached separately per language direction, preventing a previously translated Arabic page from being reused in a Persian book run.
EPUB pages with embedded cover artwork are detected for image translation.
Verified the supplied EPUB end to end in both English→Arabic and English→Persian with a deterministic local test engine (169 pages each). Initial direct Gemini verification returned HTTP 504 in this environment; see the follow-up live test below.
Regression suites: 7 issue-specific tests and 296 application checks passed. The Windows EXE was rebuilt and launched successfully on Windows 11.
Gemini connectivity follow-up
Settings now provide a separate optional HTTP(S) proxy for Gemini and an optional Gemini-native API gateway URL. The Google SDK system HTTPS_PROXY environment setting is also honored. Proxy URLs are encrypted in the Windows config.
Old loopback URLs that previously belonged to OpenAI-compatible servers are cleared from the Gemini endpoint during one-time migration.
Transient 500/502/503/504, rate-limit, and transport failures retry with backoff. If the shipped default gemini-3.8-flash remains overloaded, the app automatically retries with gemini-3.6-flash; explicitly selected custom models are not silently changed.
Live structured connection test passed with the user's configured API key: Google temporarily returned model load/rate responses for 3.8, then automatic fallback to 3.6 returned OK.
Hotfix - 2026-11-04: Fixed a data-load regression in the per-direction job store: books translated with earlier versions reopened at 0% because their translated pages were archived into a legacy bucket and never restored. Existing jobs now load with their progress intact, and data already affected is repaired automatically on first open. The executable and source archive in this release have been updated with the fix (commit ac00625).
fix: restore legacy job translations instead of hiding them · Mohammad-Hasan-Kaman/kaman-ai-translator@ac00625
Jobs created before per-direction storage hold translations without a recorded direction. switch_direction() archived those pages into legacy_translations and never restored them, so existing books...

42 0 0 1
Каталог
Каталог каналов и чатов Подборки каналов Поиск каналов Добавить канал/чат
Рейтинги
Рейтинг каналов Telegram Рейтинг чатов Telegram Рейтинг публикаций Рейтинги брендов и персон
API
API статистики API поиска публикаций API Callback
Наши каналы
@TGStat @TGStat_Chat @telepulse @TGStatAPI
Почитать
Академия TGStat Исследование Telegram 2019 Исследование Telegram 2021 Исследование Telegram 2023
Контакты
Справочный центр Поддержка Почта Вакансии
Всякая всячина
Пользовательское соглашение Политика конфиденциальности Публичная оферта
Наши боты
@TGStat_Bot @SearcheeBot @TGAlertsBot @tg_analytics_bot @TGStatChatBot