Original Summary

Modular monolith serving a LoRA fine-tuned Helsinki-NLP opus-mt-en-es translation model, training about 0.38 percent of parameters on the opus_books corpus, with sentence-aware chunking to avoid silent truncation, deterministic beam-search decoding, a thread-safe TTL cache, proxy-aware rate limiting, and a 193-test CI suite on Docker and Render.

  • 情报分类：技术学习与提效
  • 分类依据：内容涉及技术、AI、软件工具或工程实践
  • 信息来源：GitHub · AI 新项目
  • 发布时间：2026/10/14 20:00:59