docs: 将仓库代码注释统一为中文

backend/app/local_models/__init__.py

@@ -1 +1 @@
-"""Optional local inference; importing this package does not load model libraries."""
+"""可选的本地推理；导入此包不会加载模型库。"""

backend/app/local_models/catalog.py

@@ -1,4 +1,4 @@
-"""Reviewed model identities. Runtime never resolves a moving model revision."""
+"""经过审核的模型标识；运行时绝不解析浮动的模型版本。"""
 from dataclasses import asdict, dataclass
 
 

backend/app/local_models/components.py

@@ -1,4 +1,4 @@
-"""User-triggered installation of the fixed optional CUDA runtime on Windows."""
+"""用户触发在 Windows 上安装固定的可选 CUDA 运行时。"""
 import asyncio
 import json
 import os

backend/app/local_models/manager.py

@@ -1,4 +1,4 @@
-"""Explicit resumable downloads; inference itself never fetches weights."""
+"""由用户显式触发、支持断点续传的下载；推理过程本身绝不下载权重。"""
 from __future__ import annotations
 
 import asyncio

backend/app/local_models/process.py

@@ -1,4 +1,4 @@
-"""Pipe adapter for event loops without asyncio subprocess support (Windows reload)."""
+"""用于没有异步子进程支持的事件循环的管道适配器（Windows 重新加载）。"""
 from __future__ import annotations
 
 import asyncio
@@ -33,14 +33,14 @@ class _Output:
         self.limit = limit
 
     async def readline(self):
-        # Bound allocations even when the worker produces a malformed line.
+        # 即使工作线程生成格式错误的行，分配也会受到限制。
         return await asyncio.to_thread(self.pipe.readline, self.limit + 1)
 
 
 class ThreadedProcess:
     def __init__(self, args, *, env, limit, creationflags=0):
-        # Spawn synchronously so cancellation cannot leave an unowned process.
-        # Blocking pipe I/O and reaping run in threads, never on the server loop.
+        # 同步创建进程，避免取消操作留下无人管理的子进程。阻塞式管道 I/O 与进程回收在线程中执行，
+        # 不占用服务器事件循环。
         self.process = subprocess.Popen(
             args, stdin=subprocess.PIPE, stdout=subprocess.PIPE,
             stderr=subprocess.DEVNULL, env=env, creationflags=creationflags,

backend/app/local_models/protocol.py

@@ -1,4 +1,4 @@
-"""Bound embedding result frames so large notes do not exceed pipe line limits."""
+"""绑定嵌入结果帧，因此大笔记不会超出管道限制。"""
 import json
 
 

backend/app/local_models/runtime.py

@@ -1,4 +1,4 @@
-"""Bounded, cancellable model subprocesses with CPU as the default device."""
+"""有界、可取消的模型子流程，以 CPU 作为默认设备。"""
 from __future__ import annotations
 
 import asyncio
@@ -114,7 +114,7 @@ class Runtime:
             self.active[ticket] = key
             self.active_files[ticket] = {str(Path(payload[name]).resolve()) for name in ("source", "reference") if payload.get(name)}
             queue_seconds = time.monotonic() - queued_at
-            # Keep the reservation while replacing a failed CUDA process with CPU.
+            # 用 CPU 进程替换失败的 CUDA 进程时，继续占用原有资源配额。
             for device in (["cuda", "cpu"] if config.device == "cuda" else ["cpu"]):
                 started = time.monotonic()
                 diagnostics = dict(model=CATALOG[key].repository, revision=CATALOG[key].revision,

backend/app/local_models/worker.py

@@ -1,4 +1,4 @@
-"""One offline inference process. Heavy libraries stay out of the API process."""
+"""单个离线推理进程；重量级依赖不会加载到 API 进程中。"""
 from __future__ import annotations
 
 import contextlib
@@ -26,7 +26,7 @@ def decode(path, *, limit_seconds=3600, warnings=None):
                 corrupt += 1
                 if corrupt > 100:
                     raise ValueError("Too many damaged audio packets")
-                # Retain the missing packet's duration as silence so later timestamps do not shift.
+                # 将丢失数据包的持续时间保留为静音，以便后面的时间戳不会发生变化。
                 missing = max(0, round(float((packet.duration or 0) * (packet.time_base or 0)) * 16000))
                 samples += missing
                 if samples > limit_seconds * 16000:
@@ -58,7 +58,7 @@ def decode(path, *, limit_seconds=3600, warnings=None):
 
 
 def speech_regions(audio):
-    """Energy-based segmentation, not word alignment; retain original sample offsets."""
+    """基于能量的切分，而不是词对齐；保留原始样本偏移量。"""
     import numpy as np
     window = 480
     energies = [float(np.sqrt(np.mean(audio[i:i + window] ** 2))) for i in range(0, len(audio), window)]
@@ -140,7 +140,7 @@ def run(request):
                                         model_kwargs={"attn_implementation": "sdpa"})
             loaded = time.monotonic()
             result = model.encode(payload["texts"], batch_size=4, normalize_embeddings=True, show_progress_bar=False).tolist()
-            # Count the tokenizer's actual encoded input, not characters or words.
+            # 计算分词器的实际编码输入，而不是字符或单词。
             usage = {"input_tokens": int(model.tokenize(payload["texts"])["attention_mask"].sum())}
         elif operation == "transcription":
             from qwen_asr import Qwen3ASRModel
@@ -166,7 +166,7 @@ def run(request):
             loaded = time.monotonic()
             first = voice_embedding(model, decode(payload["source"]), device)
             second = voice_embedding(model, decode(payload["reference"]), device)
-            # Similarity, not a calibrated identity probability.
+            # 相似性，不是校准的身份概率。
             result = {"score": max(0.0, min(1.0, float(torch.dot(first, second))))}
         elif operation == "diarization":
             model = speaker_model(path, device)
@@ -198,14 +198,14 @@ def run(request):
 
 if __name__ == "__main__":
     request = json.loads(sys.stdin.buffer.read())
-    # Third-party progress/logging must never corrupt the protocol or leak into API errors.
+    # 第三方进度/日志记录绝不能破坏协议或泄漏到 API 错误。
     with contextlib.redirect_stdout(sys.stderr):
         try:
             response = run(request)
         except (ImportError, ModuleNotFoundError):
             response = {"error_code": "LOCAL_RUNTIME_DEPENDENCY_MISSING", "message": "本地模型运行依赖不完整，请重新运行安装脚本。"}
         except Exception as exc:
-            # Only device failures allow the host to retry once in a fresh CPU process.
+            # 只有设备故障才允许主机在新的 CPU 进程中重试一次。
             import torch
             cuda_failure = isinstance(exc, CudaInitializationError)
             cuda_oom = request.get("_actual_device") == "cuda:0" and isinstance(exc, torch.cuda.OutOfMemoryError)