Compare commits
2
Commits
c16dfb380b
..
master
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
664957b5ef | ||
|
|
1c6e01087b |
@@ -16,3 +16,11 @@
|
||||
- `LLM_API_BASE`:兼容接口地址,默认 `http://192.168.123.70:8080/v1/chat/completions`。
|
||||
- `LLM_API_KEY`:可选。
|
||||
- `LLM_MODEL`:默认模型,默认值 `default`。
|
||||
- `LLM_TIMEOUT_SECONDS`:单次请求超时,默认 `600`。
|
||||
|
||||
## 代码注释规范
|
||||
|
||||
- 本仓库所有源码(Python、TOML 等支持注释的文件)必须配有详细中文注释,说明模块职责、LLM 分批调用与 SRT 回填逻辑,确保后续维护人员可以快速理解代码工作原理。
|
||||
- 新增或修改代码时,必须同步补充或更新对应注释;不得删除已有注释。
|
||||
- 测试代码同样必须配有中文注释,说明每条测试验证的行为。
|
||||
- JSON 数据文件(`node.manifest.json`)不支持注释,字段语义以 `wov-sdk` 的 `NodeManifest` 模型注释和本文档输入/输出说明为准。
|
||||
|
||||
@@ -1,3 +1,4 @@
|
||||
# WOV LLM 节点配置:使用 uv 管理环境与依赖。
|
||||
[project]
|
||||
name = "wov-node-llm"
|
||||
version = "0.1.0"
|
||||
@@ -5,16 +6,20 @@ description = "WOV LLM translation node"
|
||||
requires-python = ">=3.11"
|
||||
dependencies = ["wov-sdk"]
|
||||
|
||||
# 本地路径依赖 wov-sdk。
|
||||
[tool.uv.sources]
|
||||
wov-sdk = { path = "../wov-sdk" }
|
||||
|
||||
# 开发依赖:pytest 与覆盖率工具。
|
||||
[dependency-groups]
|
||||
dev = ["pytest", "pytest-cov"]
|
||||
|
||||
# pytest 配置:强制 100% 行覆盖率。
|
||||
[tool.pytest.ini_options]
|
||||
testpaths = ["tests"]
|
||||
pythonpath = ["."]
|
||||
addopts = "--cov=wov_node_llm --cov-report=term-missing --cov-fail-under=100"
|
||||
|
||||
# 仅打包节点包本身。
|
||||
[tool.setuptools]
|
||||
packages = ["wov_node_llm"]
|
||||
|
||||
@@ -1,3 +1,8 @@
|
||||
"""LLM 翻译节点测试。
|
||||
|
||||
覆盖真实 HTTP 服务调用、超时配置、SRT 回填、异常与入口点启动等路径。
|
||||
"""
|
||||
|
||||
import json
|
||||
import runpy
|
||||
import threading
|
||||
@@ -10,7 +15,27 @@ from wov_node_llm.__main__ import invoke, translate_lines
|
||||
from wov_sdk.models import InvokeRequest
|
||||
|
||||
|
||||
class FakeUrlOpenResponse:
|
||||
"""模拟 urllib 响应对象,提供固定 LLM 译文内容。"""
|
||||
|
||||
def __init__(self, content: str) -> None:
|
||||
# 预编码为 Chat Completions 风格的 JSON 响应体。
|
||||
self._payload = json.dumps(
|
||||
{"choices": [{"message": {"content": content}}]}
|
||||
).encode("utf-8")
|
||||
|
||||
def read(self) -> bytes:
|
||||
return self._payload
|
||||
|
||||
def __enter__(self):
|
||||
return self
|
||||
|
||||
def __exit__(self, *args) -> bool:
|
||||
return False
|
||||
|
||||
|
||||
def _make_srt(tmp_path, count=5) -> Path:
|
||||
"""生成标准 SRT 测试文件,文本行为“原文字幕N”。"""
|
||||
lines = []
|
||||
for index in range(count):
|
||||
lines.extend(
|
||||
@@ -27,6 +52,7 @@ def _make_srt(tmp_path, count=5) -> Path:
|
||||
|
||||
|
||||
def test_translate_lines_via_fake_api(monkeypatch) -> None:
|
||||
"""验证通过真实 HTTP 服务器调用 LLM 接口并保持行顺序。"""
|
||||
class Handler(BaseHTTPRequestHandler):
|
||||
def do_POST(self) -> None:
|
||||
length = int(self.headers.get("Content-Length", "0"))
|
||||
@@ -72,6 +98,7 @@ def test_translate_lines_via_fake_api(monkeypatch) -> None:
|
||||
|
||||
|
||||
def test_translate_lines_api_error(monkeypatch) -> None:
|
||||
"""验证 LLM 接口不可用时抛出 URLError。"""
|
||||
def fail_open(request, timeout):
|
||||
raise urllib.error.URLError("api down")
|
||||
|
||||
@@ -83,7 +110,42 @@ def test_translate_lines_api_error(monkeypatch) -> None:
|
||||
pass
|
||||
|
||||
|
||||
def test_translate_lines_default_timeout(monkeypatch) -> None:
|
||||
"""验证未配置超时时使用默认 600 秒。"""
|
||||
captured = {}
|
||||
|
||||
def fake_open(request, timeout):
|
||||
captured["timeout"] = timeout
|
||||
return FakeUrlOpenResponse("译文一")
|
||||
|
||||
monkeypatch.setattr("wov_node_llm.__main__.urllib.request.urlopen", fake_open)
|
||||
monkeypatch.setenv("LLM_API_BASE", "http://fake/v1/chat/completions")
|
||||
|
||||
result = translate_lines(["一"], {})
|
||||
|
||||
assert result == ["译文一"]
|
||||
assert captured["timeout"] == 600
|
||||
|
||||
|
||||
def test_translate_lines_env_timeout(monkeypatch) -> None:
|
||||
"""验证 LLM_TIMEOUT_SECONDS 环境变量可覆盖超时。"""
|
||||
captured = {}
|
||||
|
||||
def fake_open(request, timeout):
|
||||
captured["timeout"] = timeout
|
||||
return FakeUrlOpenResponse("译文一")
|
||||
|
||||
monkeypatch.setattr("wov_node_llm.__main__.urllib.request.urlopen", fake_open)
|
||||
monkeypatch.setenv("LLM_API_BASE", "http://fake/v1/chat/completions")
|
||||
monkeypatch.setenv("LLM_TIMEOUT_SECONDS", "45")
|
||||
|
||||
translate_lines(["一"], {})
|
||||
|
||||
assert captured["timeout"] == 45
|
||||
|
||||
|
||||
def test_invoke_success(tmp_path, monkeypatch) -> None:
|
||||
"""验证成功调用会把译文回填到 SRT 并输出 cn.srt。"""
|
||||
source = _make_srt(tmp_path)
|
||||
|
||||
def fake_translate(lines, params):
|
||||
@@ -105,6 +167,7 @@ def test_invoke_success(tmp_path, monkeypatch) -> None:
|
||||
|
||||
|
||||
def test_invoke_pads_short_translation(tmp_path, monkeypatch) -> None:
|
||||
"""验证译文行数不足时用空行补齐,保持 SRT 结构完整。"""
|
||||
source = _make_srt(tmp_path, count=3)
|
||||
monkeypatch.setattr(
|
||||
"wov_node_llm.__main__.translate_lines",
|
||||
@@ -122,6 +185,7 @@ def test_invoke_pads_short_translation(tmp_path, monkeypatch) -> None:
|
||||
|
||||
|
||||
def test_invoke_missing_input(tmp_path) -> None:
|
||||
"""验证缺少 srt_uri 时返回失败。"""
|
||||
response = invoke(
|
||||
InvokeRequest(
|
||||
run_id="run_3",
|
||||
@@ -134,6 +198,7 @@ def test_invoke_missing_input(tmp_path) -> None:
|
||||
|
||||
|
||||
def test_invoke_missing_file(tmp_path) -> None:
|
||||
"""验证 SRT 文件不存在时返回失败。"""
|
||||
response = invoke(
|
||||
InvokeRequest(
|
||||
run_id="run_4",
|
||||
@@ -146,6 +211,7 @@ def test_invoke_missing_file(tmp_path) -> None:
|
||||
|
||||
|
||||
def test_entrypoint(monkeypatch) -> None:
|
||||
"""验证 python -m wov_node_llm 会加载 llm-translate manifest。"""
|
||||
module_path = Path(__file__).resolve().parent.parent / "wov_node_llm" / "__main__.py"
|
||||
captured = {}
|
||||
|
||||
|
||||
@@ -1 +1,5 @@
|
||||
"""WOV LLM translation node."""
|
||||
"""WOV LLM 字幕翻译节点。
|
||||
|
||||
调用 OpenAI 兼容接口,把 ASR 产出的日文 SRT 字幕逐批翻译为目标语言,
|
||||
同时保持 SRT 的序号与时间轴结构不变。
|
||||
"""
|
||||
|
||||
@@ -1,3 +1,9 @@
|
||||
"""LLM 翻译节点入口。
|
||||
|
||||
通过标准节点 HTTP 服务接收 SRT,提取纯文本行分批调用 LLM,再把译文回填到
|
||||
原 SRT 结构并输出 cn.srt。
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
@@ -9,22 +15,29 @@ from pathlib import Path
|
||||
from wov_sdk.models import InvokeRequest, InvokeResponse, NodeManifest
|
||||
from wov_sdk.server import run_node
|
||||
|
||||
# 单次 LLM 请求携带的字幕行数;过大会超出模型上下文,过小则请求次数过多。
|
||||
CHUNK_SIZE = 20
|
||||
|
||||
|
||||
def translate_lines(lines: list[str], params: dict) -> list[str]:
|
||||
"""分批调用 LLM 翻译纯文本行,返回顺序一致的译文列表。"""
|
||||
# 接口地址、Key 和模型均可通过环境变量配置,默认指向内网兼容接口。
|
||||
api_base = os.getenv(
|
||||
"LLM_API_BASE",
|
||||
"http://192.168.123.70:8080/v1/chat/completions",
|
||||
)
|
||||
api_key = os.getenv("LLM_API_KEY", "")
|
||||
# 单次请求超时可配置,长文本翻译场景下需要放宽。
|
||||
request_timeout = float(os.getenv("LLM_TIMEOUT_SECONDS", "600"))
|
||||
model = str(params.get("model") or os.getenv("LLM_MODEL", "default"))
|
||||
target_language = str(params.get("target_language", "zh-CN"))
|
||||
# 系统提示词约束模型只输出译文,保证行数和顺序可回填。
|
||||
system_prompt = (
|
||||
"你是专业字幕翻译。将用户提供的日文字幕翻译为"
|
||||
f"{target_language}。只返回译文,保持行数和顺序,不要添加解释。"
|
||||
)
|
||||
translated: list[str] = []
|
||||
# 按 CHUNK_SIZE 分批发送,避免单次请求超过模型上下文限制。
|
||||
for start in range(0, len(lines), CHUNK_SIZE):
|
||||
chunk = lines[start : start + CHUNK_SIZE]
|
||||
body = {
|
||||
@@ -35,6 +48,7 @@ def translate_lines(lines: list[str], params: dict) -> list[str]:
|
||||
],
|
||||
}
|
||||
headers = {"Content-Type": "application/json"}
|
||||
# 配置了 Key 时附带 Bearer 鉴权头。
|
||||
if api_key:
|
||||
headers["Authorization"] = f"Bearer {api_key}"
|
||||
request = urllib.request.Request(
|
||||
@@ -43,9 +57,11 @@ def translate_lines(lines: list[str], params: dict) -> list[str]:
|
||||
headers=headers,
|
||||
method="POST",
|
||||
)
|
||||
with urllib.request.urlopen(request, timeout=120) as response:
|
||||
with urllib.request.urlopen(request, timeout=request_timeout) as response:
|
||||
payload = json.loads(response.read().decode("utf-8"))
|
||||
# 兼容 OpenAI Chat Completions 响应格式,取第一条消息内容。
|
||||
content = payload["choices"][0]["message"]["content"]
|
||||
# 忽略空行,保证译文列表与输入行一一对应。
|
||||
translated.extend(
|
||||
[line.strip() for line in content.splitlines() if line.strip()]
|
||||
)
|
||||
@@ -53,6 +69,7 @@ def translate_lines(lines: list[str], params: dict) -> list[str]:
|
||||
|
||||
|
||||
def invoke(request: InvokeRequest) -> InvokeResponse:
|
||||
"""翻译 SRT 文件中的字幕文本,输出 cn.srt。"""
|
||||
srt_uri = request.inputs.get("srt_uri")
|
||||
if not srt_uri:
|
||||
return InvokeResponse(status="failed", error="srt_uri is required")
|
||||
@@ -61,23 +78,28 @@ def invoke(request: InvokeRequest) -> InvokeResponse:
|
||||
if not srt_path.is_file():
|
||||
return InvokeResponse(status="failed", error="srt file not found")
|
||||
|
||||
# 标准 SRT 每 4 行一组:序号、时间轴、文本、空行;文本位于第 3 行。
|
||||
lines = srt_path.read_text(encoding="utf-8").splitlines()
|
||||
text_indices = list(range(2, len(lines), 4))
|
||||
source_lines = [lines[index] for index in text_indices]
|
||||
translated_lines = translate_lines(source_lines, request.params)
|
||||
# 防止模型返回行数偏差:多出的截断,缺少的用空串补齐。
|
||||
translated_lines = translated_lines[: len(source_lines)]
|
||||
translated_lines += [""] * max(0, len(source_lines) - len(translated_lines))
|
||||
# 只替换文本行,序号、时间轴和空行保持不变。
|
||||
for index, text_index in enumerate(text_indices):
|
||||
lines[text_index] = translated_lines[index]
|
||||
|
||||
output_dir = Path(request.output_dir)
|
||||
output_dir.mkdir(parents=True, exist_ok=True)
|
||||
output_path = output_dir / "cn.srt"
|
||||
# 末尾补一个换行,让文件满足常见文本工具习惯。
|
||||
output_path.write_text("\n".join(lines) + "\n", encoding="utf-8")
|
||||
return InvokeResponse(status="completed", outputs={"cn_srt_uri": str(output_path)})
|
||||
|
||||
|
||||
def main() -> None:
|
||||
"""加载节点清单并以本模块的 invoke 处理器启动服务。"""
|
||||
manifest_path = Path(__file__).resolve().parent.parent / "node.manifest.json"
|
||||
with open(manifest_path, "r", encoding="utf-8") as f:
|
||||
manifest = NodeManifest.from_dict(json.load(f))
|
||||
|
||||
Reference in New Issue
Block a user