[DOC] Fix wrong CN translate (vllm-project#11652)

Fix wrong CN translate

- vLLM version: v0.23.0
- vLLM main:
vllm-project/vllm@1f486d9

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

Author: wangxiyuan

.github/workflows/scripts/detect_po_changes.py

@@ -280,6 +280,38 @@ def _po_entry_block(msgid: str) -> str:
     return "\n".join(parts)
 
 
+def _dedup_po_entries(po: POFile) -> int:
+    """Remove duplicate entries from *po*, keeping only the first.
+
+    Covers two cases:
+    - Duplicate msgid values (same paragraph extracted multiple times).
+    - Duplicate empty msgid (``msgid ""``) — only the first (which carries
+      the PO header metadata like ``Project-Id-Version``) is kept; any
+      additional bare ``msgid ""`` entries are removed.  Without this,
+      ``polib`` may shuffle the bare entry's position relative to the
+      header block on every save, causing spurious diffs.
+
+    Returns the number of entries removed.
+    """
+    seen: set[str] = set()
+    removed = 0
+    empty_seen = False
+    for entry in list(po):
+        if not entry.msgid:
+            if empty_seen:
+                po.remove(entry)
+                removed += 1
+            else:
+                empty_seen = True
+            continue
+        if entry.msgid in seen:
+            po.remove(entry)
+            removed += 1
+        else:
+            seen.add(entry.msgid)
+    return removed
+
+
 def process_file(source_path: Path, dry_run: bool = False, force: bool = False) -> bool:
     """Create or update the .po file for *source_path*.
 
@@ -310,6 +342,14 @@ def process_file(source_path: Path, dry_run: bool = False, force: bool = False)
 
     # Incremental update: detect new, removed, and modified paragraphs.
     po = pofile(str(po_path))
+
+    # Deduplicate existing entries — keep only the first occurrence of each
+    # msgid; mark duplicates as obsolete.  Without this step, repeated
+    # invocations on files that already contain duplicate msgid entries
+    # (e.g. from prior force rebuilds or buggy incremental runs) would
+    # keep appending more copies.
+    _dedup_po_entries(po)
+
     entries_by_msgid: dict[str, POEntry] = {}
     for entry in po:
         if entry.msgid:
@@ -319,8 +359,25 @@ def process_file(source_path: Path, dry_run: bool = False, force: bool = False)
     modified_count = 0
     removed_count = 0
 
+    # Track how many times each paragraph appears in the source so that
+    # repeated paragraphs (e.g. "Offline example:") consume one PO entry
+    # each without creating spurious "new" entries.
+    para_counts: dict[str, int] = {}
+    for para in paragraphs:
+        para_counts[para] = para_counts.get(para, 0) + 1
+
     for para in paragraphs:
-        if para in entries_by_msgid:
+        remaining = para_counts.get(para, 0)
+        if para in entries_by_msgid and remaining > 0:
+            del entries_by_msgid[para]
+            para_counts[para] = remaining - 1
+            continue
+        # If the source has more occurrences of *para* than the PO has
+        # entries, we need to create additional entries.
+        if remaining > 0:
+            para_counts[para] = remaining - 1
+            po.append(_build_po_entry(para))
+            new_count += 1
             continue
         # Check if this is a modification of an existing paragraph
         # (similar msgid that got updated in the source).
@@ -357,6 +414,13 @@ def process_file(source_path: Path, dry_run: bool = False, force: bool = False)
         return True
 
     po.save(str(po_path))
+    # Work around polib 1.2.0 unstable header serialization (the bare
+    # ``msgid ""``/``msgstr ""`` block drifts relative to the metadata
+    # header block on consecutive saves).  Loading and saving a second
+    # time stabilizes the output — polib's own round-trip is then
+    # deterministic.
+    po2 = pofile(str(po_path))
+    po2.save(str(po_path))
     parts = []
     if new_count:
         parts.append(f"+{new_count} new")

.github/workflows/scripts/po_translate.py

@@ -57,9 +57,16 @@
 
 --- WHAT TO PRESERVE (keep EXACTLY as in msgid) ---
 4. All format specifiers: %s, %d, %f, {{}}, {{{{}}}}, {{name}}, etc.
-5. All markdown syntax: **bold**, *italic*, `inline code`, ```code blocks```,
-   [links](urls), ![images](urls), # headings, - lists, 1. ordered lists,
-   > blockquotes, | tables, --- horizontal rules.
+5. All markdown syntax MUST be preserved exactly:
+   - Headings: "#", "##", "###", etc. at the start of a line
+   - Bold/italic: **bold**, *italic*
+   - Inline code: `code`
+   - Code blocks: ```code blocks```
+   - Links: [text](url), ![images](url)
+   - List markers: "- ", "* ", "1. ", "2. " at the start of each list item
+   - Blockquotes: "> "
+   - Tables: | columns |, separator rows (|:---:|)
+   - Horizontal rules: ---
 6. HTML tags and attributes: <div>, <a href>, <img>, etc.
 7. Environment variables, file paths, command names, code identifiers.
 8. Proper nouns: person names, contributor names, author names, company names,
@@ -103,7 +110,10 @@
 18. For markdown links [text](url): translate the display text in [] but keep
     the URL in () exactly as-is.
     Example: [Quick Start](quick_start.md) → [快速开始](quick_start.md)
-19. For headings (# Title): translate the heading text.
+19. For headings: translate the heading text but KEEP the "#" / "##" / "###"
+    prefix exactly as in the msgid.
+    Example: msgid "# Quick Start" → msgstr "# 快速开始"
+    Example: msgid "## Overview" → msgstr "## 概述"
 20. DO NOT add "#, fuzzy" markers.
 21. If a msgid is purely structural (symbols, code, file paths only), copy it
     verbatim to msgstr.

docs/source/locale/zh_CN/LC_MESSAGES/community/slash-commands.po

@@ -39,9 +39,6 @@ msgid ""
 "automatically routed to the appropriate NPU runner based on the test path."
 msgstr "运行 `tests/e2e/pull_request/` 下的特定 E2E 测试。测试会根据路径自动路由到相应的 NPU 运行器。"
 
-msgid "**Examples:**"
-msgstr "**示例：**"
-
 msgid "**Routing rules** (matched in order):"
 msgstr "**路由规则**（按顺序匹配）："
 
@@ -88,9 +85,6 @@ msgstr ""
 "`schedule_nightly_test_a2.yaml` 和 `schedule_nightly_test_a3.yaml` 中定义的 "
 "`test_config.name` 条目。"
 
-msgid "**Usage:**"
-msgstr "**用法：**"
-
 msgid ""
 "| Syntax | Scope |\n"
 "|---|---|\n"
@@ -152,9 +146,6 @@ msgstr ""
 "`deepseek-r1-0528-w8a8`, `kimi-k2-thinking`, "
 "`qwen3-vl-235b-a22b-instruct-w8a8`, `custom-multi-ops`, ..."
 
-msgid "**Examples:**"
-msgstr "**示例：**"
-
 msgid ""
 "This triggers `workflow_dispatch` on both `schedule_nightly_test_a2.yaml` "
 "and `schedule_nightly_test_a3.yaml`."
@@ -170,9 +161,6 @@ msgid ""
 "PR. This is useful for backporting fixes to release branches."
 msgstr "将一个 PR 的提交 cherry-pick 到指定的目标分支并创建一个新的 PR。这对于将修复反向移植到发布分支非常有用。"
 
-msgid "**Usage:**"
-msgstr "**用法：**"
-
 msgid ""
 "| Syntax | Description |\n"
 "|---|---|\n"
@@ -182,9 +170,6 @@ msgstr ""
 "|---|---|\n"
 "| `/cherry-pick <target_branch>` | Cherry-pick 到指定分支 |"
 
-msgid "**Examples:**"
-msgstr "**示例：**"
-
 msgid ""
 "A new PR will be created with the title format `[Cherry-pick] "
 "<original_title> (from #<PR_NUMBER>)` and a body linking back to the "
@@ -304,3 +289,18 @@ msgid ""
 msgstr ""
 "权限通过 GitHub API (`repos/{owner}/{repo}/collaborators/{user}/permission`) "
 "进行验证。"
+
+msgid "**Examples:**"
+msgstr ""
+
+msgid "**Usage:**"
+msgstr ""
+
+msgid "**Examples:**"
+msgstr ""
+
+msgid "**Usage:**"
+msgstr ""
+
+msgid "**Examples:**"
+msgstr ""

docs/source/locale/zh_CN/LC_MESSAGES/developer_guide/Design_Documents/ModelRunner_prepare_inputs.po

@@ -170,9 +170,6 @@ msgstr "这些假设条件在启动vLLM时配置。它们并非固定不变，
 msgid "### Step 1: All requests in the prefill phase"
 msgstr "### 步骤1：所有请求处于预填充阶段"
 
-msgid "#### Obtain inputs"
-msgstr "#### 获取输入"
-
 msgid ""
 "As the maximum number of tokens that can be scheduled is 10, the scheduled "
 "tokens of each request can be represented as `{'0': 3, '1': 2, '2': 5}`. "
@@ -272,9 +269,6 @@ msgstr ""
 "- `Input IDs` = `[T_0_0, T_0_1, T_0_2, T_1_0, T_1_1, T_2_0, T_2_1, T_3_2, "
 "T_3_3, T_3_4]`"
 
-msgid "#### Build inputs attention metadata"
-msgstr "#### 构建输入注意力元数据"
-
 msgid ""
 "In the current **Block Table**, we use the first block (i.e. block_0) to "
 "mark the unused block. The shape of the block is `(max num request, max "
@@ -471,3 +465,9 @@ msgid ""
 "model forwarding. If you have any good ideas, you are welcome to contribute "
 "to us."
 msgstr "希望本文档能帮助你更好地理解 vLLM 如何为模型前向传播准备输入。如果你有任何好的想法，欢迎向我们贡献。"
+
+msgid "#### Obtain inputs"
+msgstr ""
+
+msgid "#### Build inputs attention metadata"
+msgstr ""