Skip to content

7.4 billion tokens per month. 34 free LLM providers. 635 free model endpoints. All behind one /v1 endpoint, plus any custom OpenAI-compatible endpoint. Smart routing, automatic failover, encrypted keys. Personal experimentation only.

TypeScript

docs(i18n): pin the zh glossary, land the two agreed zh-CN fixes (#683)

Tashfeenc8be5a68b2fc
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 4a5c1ff0..4acd0452 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -32,6 +32,13 @@ npm run db:migration:up
npm run db:migration:down
```
+## Translations
+
+The dashboard ships 60 locales. `en.json` is the source of truth and every other file mirrors
+its keys, so run `npm run check:i18n` from `client/` before opening a PR. See
+[docs/translating.md](docs/translating.md) for the full rules and the settled Chinese
+terminology.
+
## AI and LLM-assisted contributions
LLM-assisted PRs are welcome. A lot of this codebase is itself built that way, so there is no stigma here. The bar is the same as for any other PR: you are responsible for what you submit.
diff --git a/client/src/i18n/locales/zh-CN.json b/client/src/i18n/locales/zh-CN.json
index 0f1465bb..dd4ca3f0 100644
--- a/client/src/i18n/locales/zh-CN.json
+++ b/client/src/i18n/locales/zh-CN.json
@@ -86,7 +86,7 @@
"generalDescription": "此仪表盘的语言与外观。",
"advancedDescription": "压缩流程的阈值与安全开关。",
"compressionTitle": "提示词压缩",
- "compressionDescription": "在路由前减小请求数据量,同时保留代码、错误信息、路径、数字和工具定义。",
+ "compressionDescription": "在路由前减小请求大小,同时保留代码、错误信息、路径、数字和工具定义。",
"compressionHelp": "在请求发送给提供方之前对其进行改写,使其占用更少的词元。代码、错误信息、文件路径和数字受到保护,任何会破坏它们的改写都会被丢弃。",
"compressionMode": "模式",
"compressionModeHelp": "关闭时原样发送请求。无损仅移除重复和多余的文本。标准还会缩短工具输出。激进则会额外压缩较早的对话轮次。",
@@ -171,7 +171,7 @@
"codeSnippetHeading": "使用该模型",
"modelNotFound": "该模型不可用,或尚未配置任何提供方。",
"providerIdsHeading": "各提供方的模型 ID",
- "providerIdsHint": "使用统一模型 ID 可在所有提供方之间自动故障转移;使用某个提供方专属的 ID 则指定经由该提供方。",
+ "providerIdsHint": "使用统一模型 ID 可在所有提供方之间自动故障转移;使用某个提供方专属的 ID 则会将请求锁定到该提供方。",
"copyModelName": "复制模型名称",
"settingsHeading": "提供方设置",
"settingsHint": "对目录模型的本地修改会在目录同步后保留。已删除的目录模型会保持删除状态,直到你通过配置重新添加。",
diff --git a/docs/translating.md b/docs/translating.md
new file mode 100644
index 00000000..6776f51a
--- /dev/null
+++ b/docs/translating.md
@@ -0,0 +1,71 @@
+# Translating
+
+The dashboard ships 60 locales under `client/src/i18n/locales/`. `en.json` is the source of
+truth: every other file mirrors its key structure exactly.
+
+## Before you open a PR
+
+- Never add or remove keys in a locale file on its own. If a string needs to exist, it starts
+ in `en.json` and every locale follows.
+- Run the validator from `client/`:
+
+ ```bash
+ npm run check:i18n
+ ```
+
+ It checks all 60 locales for key parity, value types, and matching `{placeholder}` names.
+ It does not check whether the translation is any good, so read your diff too.
+- Keep a PR to one locale where you can. A 60-file diff is very hard to review.
+- Translate the meaning, not the words. If an English string is ambiguous, look at where it
+ renders before guessing.
+
+## Chinese terminology
+
+The zh-CN terms below were settled in [#669](https://github.com/tashfeenahmed/freellmapi/pull/669)
+after a long review. They are written down so the next pass does not relitigate them. If you
+want to change one, bring a reason that is stronger than preference: a standards body, a
+reference product, or a real ambiguity in the current wording.
+
+| English | zh-CN | Notes |
+| --- | --- | --- |
+| Provider | 提供方 | Not 提供商. Custom endpoints, local Ollama, and community-run instances are not vendors. |
+| Token (LLM) | 词元 | The [national standard term](https://termonline.cn). See the note below. |
+| Token (auth, API keys, URL tokens) | 令牌 | Never 词元. These are credentials, not word units. |
+| Coding | 编程 | Not 编码, which reads as "encoding". |
+| Export | 导出 | Not 出口, which is the shipping sense. |
+| Request size | 请求大小 | Not 请求数据量, which reads as an aggregate across many requests. |
+| Request body | 请求正文 | What Google Cloud and Cloudflare use in their own Chinese docs. 请求体 is commoner in speech and is not wrong, but use 请求正文 here. |
+| Revoke | 撤销 | Not 废除, which is closer to repealing a law. |
+| Playground | 试验台 | It is a bench for testing models. Not 试玩台, where 试玩 suggests a game demo. |
+| Balance (the custom preset) | 权衡 | The numeric sliders themselves are 权重. |
+
+The strategy presets are comparative, so they keep 最: 最快, 最稳定, 最智能.
+
+Punctuation is full-width (,。:;), and Chinese terms do not take Latin-style spaces around
+them. Latin words and numbers embedded in Chinese text do take a space on each side, as in
+`API 令牌`.
+
+### On 词元
+
+This one has real dissent on the record and is worth understanding before you touch it.
+词元 is the term the national sci-tech terminology committee publishes, and it is unambiguous
+next to 令牌 for auth credentials, which the file previously confused. Against that, most
+Chinese AI products still show `Token` in Latin in their own interfaces, so 词元 will look
+formal to some users.
+
+It is decided, not unnoticed. Changing it means changing roughly 25 strings, so raise it as an
+issue first rather than folding it into an unrelated PR.
+
+## zh-TW is not zh-CN
+
+zh-TW diverges on purpose and should not be synced to zh-CN term for term:
+
+| English | zh-TW |
+| --- | --- |
+| Token | `Token`, left in Latin |
+| Provider | 提供者 |
+| Playground | 遊樂場 |
+
+Taiwanese technical writing keeps far more Latin than mainland writing does. If you are
+updating one of the two files, check the other for the same string, but match the local
+convention rather than making them identical.