vfx (@freebsdfx) 在 测试了一下本地翻译需要的模型最小参数. 中发帖
用的是ollama+gemma,测试下来12b以上表达算比较流畅(量化后要求显存10G左右),1b和4b对于翻译的准确性来说,还是有点不够。
小说原文:
“Or else it’s the house. The house is haunted—or bewitched or something … I see things that have happened there—or else I see things that are going to happen there—and that would be worse. Perhaps a woman called Helen isgoing to be murdered there … Only I don’t see if it’s the house that’s haunted why I should see these...