mirror of
https://github.com/JefferyHcool/BiliNote.git
synced 2026-06-13 03:30:10 +08:00
41f17592c2
- whisper: model.bin 截断/损坏时删目录重下重试一次，修「Unable to
open file model.bin」死循环；mlx 同样按 config.json 判完整性
- /generate_note 加就绪门禁：本地转写引擎模型没下好直接拦截，返回
reason=transcriber_model_not_ready，不让任务静默卡在首次下载
- 全局代理：新增 ProxyConfigManager（JSON 配置 + HTTP_PROXY env 兜底）
+ build_openai_client，统一注入代理到 LLM/Groq 客户端；yt-dlp 与
youtube-transcript-api 也走代理
- build_openai_client 校验 api_key 非空，空 key 给「xxx 的 API Key
未配置」而不是天书般的 Illegal header value b'Bearer '
- universal_gpt: 模型拒绝自定义 temperature（o1/o3/gpt-5 系列）时
就地去掉参数重试，不消耗重试预算
- connect_test 改用真实 chat completion 而非 /v1/models 探测
- main.py: lifespan 拆 [startup 1/5..5/5] 分段日志 + 异常清晰定位
- /sys_health 重构为结构化返回 {backend,ffmpeg,db,whisper_model}
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
385 lines
13 KiB
Python
385 lines
13 KiB
Python
import os
|
||
import platform
|
||
from pathlib import Path
|
||
|
||
from fastapi import APIRouter, HTTPException, BackgroundTasks
|
||
from pydantic import BaseModel
|
||
from typing import Optional
|
||
from app.utils.response import ResponseWrapper as R
|
||
from app.utils.logger import get_logger
|
||
from app.utils.path_helper import get_model_dir
|
||
|
||
from app.services.cookie_manager import CookieConfigManager
|
||
from app.services.transcriber_config_manager import TranscriberConfigManager
|
||
from ffmpeg_helper import ensure_ffmpeg_or_raise
|
||
|
||
logger = get_logger(__name__)
|
||
|
||
router = APIRouter()
|
||
cookie_manager = CookieConfigManager()
|
||
transcriber_config_manager = TranscriberConfigManager()
|
||
|
||
|
||
class CookieUpdateRequest(BaseModel):
|
||
platform: str
|
||
cookie: str
|
||
|
||
|
||
@router.get("/get_downloader_cookie/{platform}")
|
||
def get_cookie(platform: str):
|
||
cookie = cookie_manager.get(platform)
|
||
if not cookie:
|
||
return R.success(msg='未找到Cookies')
|
||
return R.success(
|
||
data={"platform": platform, "cookie": cookie}
|
||
)
|
||
|
||
|
||
@router.post("/update_downloader_cookie")
|
||
def update_cookie(data: CookieUpdateRequest):
|
||
cookie_manager.set(data.platform, data.cookie)
|
||
return R.success(
|
||
|
||
)
|
||
|
||
class TranscriberConfigRequest(BaseModel):
|
||
transcriber_type: str
|
||
whisper_model_size: Optional[str] = None
|
||
|
||
|
||
AVAILABLE_TRANSCRIBER_TYPES = [
|
||
{"value": "fast-whisper", "label": "Faster Whisper（本地）"},
|
||
{"value": "bcut", "label": "必剪（在线）"},
|
||
{"value": "kuaishou", "label": "快手（在线）"},
|
||
{"value": "groq", "label": "Groq（在线）"},
|
||
{"value": "mlx-whisper", "label": "MLX Whisper（仅macOS）"},
|
||
]
|
||
|
||
WHISPER_MODEL_SIZES = ["tiny", "base", "small", "medium", "large-v3", "large-v3-turbo"]
|
||
|
||
|
||
@router.get("/transcriber_config")
|
||
def get_transcriber_config():
|
||
from app.transcriber.transcriber_provider import MLX_WHISPER_AVAILABLE
|
||
|
||
config = transcriber_config_manager.get_config()
|
||
return R.success(data={
|
||
**config,
|
||
"available_types": AVAILABLE_TRANSCRIBER_TYPES,
|
||
"whisper_model_sizes": WHISPER_MODEL_SIZES,
|
||
"mlx_whisper_available": MLX_WHISPER_AVAILABLE,
|
||
})
|
||
|
||
|
||
@router.post("/transcriber_config")
|
||
def update_transcriber_config(data: TranscriberConfigRequest):
|
||
config = transcriber_config_manager.update_config(
|
||
transcriber_type=data.transcriber_type,
|
||
whisper_model_size=data.whisper_model_size,
|
||
)
|
||
return R.success(data=config)
|
||
|
||
|
||
# ---- 全局代理配置（作用于 LLM API + 转写 API + yt-dlp 下载）----
|
||
|
||
class ProxyConfigRequest(BaseModel):
|
||
enabled: bool
|
||
url: Optional[str] = None
|
||
|
||
|
||
@router.get("/proxy_config")
|
||
def get_proxy_config():
|
||
from app.services.proxy_config_manager import ProxyConfigManager
|
||
mgr = ProxyConfigManager()
|
||
cfg = mgr.get_config()
|
||
# effective 给前端展示「当前实际生效的代理」——可能来自配置，也可能来自 env 兜底
|
||
return R.success(data={
|
||
**cfg,
|
||
"effective": mgr.get_proxy_url() or "",
|
||
})
|
||
|
||
|
||
@router.post("/proxy_config")
|
||
def update_proxy_config(data: ProxyConfigRequest):
|
||
from app.services.proxy_config_manager import ProxyConfigManager
|
||
mgr = ProxyConfigManager()
|
||
cfg = mgr.update_config(enabled=data.enabled, url=data.url)
|
||
return R.success(data={
|
||
**cfg,
|
||
"effective": mgr.get_proxy_url() or "",
|
||
})
|
||
|
||
|
||
# ---- Whisper 模型下载状态 & 下载触发 ----
|
||
|
||
# 用于跟踪正在进行的下载任务
|
||
_downloading: dict[str, str] = {} # model_size -> status ("downloading" | "done" | "failed")
|
||
|
||
|
||
def _check_whisper_model_exists(model_size: str, subdir: str = "whisper") -> bool:
|
||
"""检查指定 whisper 模型是否已下载完整到本地。
|
||
|
||
必须 model.bin 落盘才算完成，仅有空目录或半成品不能算「已下载」——
|
||
否则监控页会显示绿勾但加载时报「Unable to open file 'model.bin'」。
|
||
"""
|
||
model_dir = get_model_dir(subdir)
|
||
model_path = os.path.join(model_dir, f"whisper-{model_size}")
|
||
return (Path(model_path) / "model.bin").exists()
|
||
|
||
|
||
def _check_mlx_whisper_model_exists(model_size: str) -> bool:
|
||
"""检查 mlx-whisper 模型是否已下载完整到本地。
|
||
|
||
与 fast-whisper 的目录布局不同：mlx 模型按 HuggingFace repo_id
|
||
（如 mlx-community/whisper-tiny-mlx）落盘，且没有 model.bin，
|
||
用 config.json 作为「下载完成」的判据，和 mlx_whisper_transcriber.py 保持一致。
|
||
"""
|
||
try:
|
||
from app.transcriber.mlx_whisper_transcriber import MLX_MODEL_MAP
|
||
except Exception:
|
||
return False
|
||
repo_id = MLX_MODEL_MAP.get(model_size)
|
||
if not repo_id:
|
||
return False
|
||
model_dir = get_model_dir("mlx-whisper")
|
||
model_path = os.path.join(model_dir, repo_id)
|
||
return (Path(model_path) / "config.json").exists()
|
||
|
||
|
||
@router.get("/transcriber_models_status")
|
||
def get_transcriber_models_status():
|
||
"""返回所有 whisper 模型的下载状态。"""
|
||
statuses = []
|
||
for size in WHISPER_MODEL_SIZES:
|
||
downloaded = _check_whisper_model_exists(size, "whisper")
|
||
download_status = _downloading.get(size)
|
||
statuses.append({
|
||
"model_size": size,
|
||
"downloaded": downloaded,
|
||
"downloading": download_status == "downloading",
|
||
})
|
||
|
||
# 也检查 mlx-whisper（仅 macOS）
|
||
mlx_available = platform.system() == "Darwin"
|
||
mlx_statuses = []
|
||
if mlx_available:
|
||
from app.transcriber.mlx_whisper_transcriber import MLX_MODEL_MAP
|
||
for size in WHISPER_MODEL_SIZES:
|
||
mlx_key = f"mlx-{size}"
|
||
repo_id = MLX_MODEL_MAP.get(size)
|
||
# 用 config.json 判定，和 _check_mlx_whisper_model_exists / 加载逻辑保持一致
|
||
downloaded = _check_mlx_whisper_model_exists(size)
|
||
mlx_statuses.append({
|
||
"model_size": size,
|
||
"downloaded": downloaded,
|
||
"downloading": _downloading.get(mlx_key) == "downloading",
|
||
"available": repo_id is not None,
|
||
})
|
||
|
||
return R.success(data={
|
||
"whisper": statuses,
|
||
"mlx_whisper": mlx_statuses,
|
||
"mlx_available": mlx_available,
|
||
})
|
||
|
||
|
||
class ModelDownloadRequest(BaseModel):
|
||
model_size: str
|
||
transcriber_type: str = "fast-whisper" # "fast-whisper" 或 "mlx-whisper"
|
||
|
||
|
||
def _do_download_whisper(model_size: str):
|
||
"""后台下载 faster-whisper 模型。"""
|
||
from app.transcriber.whisper import MODEL_MAP
|
||
from modelscope import snapshot_download
|
||
|
||
try:
|
||
_downloading[model_size] = "downloading"
|
||
model_dir = get_model_dir("whisper")
|
||
model_path = os.path.join(model_dir, f"whisper-{model_size}")
|
||
# 用 model.bin 判定而非目录存在：半成品目录不能算「已下载」
|
||
if (Path(model_path) / "model.bin").exists():
|
||
_downloading[model_size] = "done"
|
||
return
|
||
repo_id = MODEL_MAP.get(model_size)
|
||
if not repo_id:
|
||
_downloading[model_size] = "failed"
|
||
return
|
||
logger.info(f"开始下载 whisper 模型: {model_size}")
|
||
snapshot_download(repo_id, local_dir=model_path)
|
||
logger.info(f"whisper 模型下载完成: {model_size}")
|
||
_downloading[model_size] = "done"
|
||
except Exception as e:
|
||
logger.error(f"whisper 模型下载失败: {model_size}, {e}")
|
||
_downloading[model_size] = "failed"
|
||
|
||
|
||
def _do_download_mlx_whisper(model_size: str):
|
||
"""后台下载 mlx-whisper 模型。"""
|
||
key = f"mlx-{model_size}"
|
||
try:
|
||
_downloading[key] = "downloading"
|
||
from huggingface_hub import snapshot_download as hf_download
|
||
from app.transcriber.mlx_whisper_transcriber import resolve_mlx_repo_id
|
||
|
||
try:
|
||
repo_id = resolve_mlx_repo_id(model_size)
|
||
except ValueError as e:
|
||
logger.error(str(e))
|
||
_downloading[key] = "failed"
|
||
return
|
||
|
||
model_dir = get_model_dir("mlx-whisper")
|
||
model_path = os.path.join(model_dir, repo_id)
|
||
# 用 config.json 判定而非目录存在：半成品目录不能算「已下载」
|
||
if (Path(model_path) / "config.json").exists():
|
||
_downloading[key] = "done"
|
||
return
|
||
logger.info(f"开始下载 mlx-whisper 模型: {model_size} ← {repo_id}")
|
||
hf_download(repo_id, local_dir=model_path, local_dir_use_symlinks=False)
|
||
logger.info(f"mlx-whisper 模型下载完成: {model_size}")
|
||
_downloading[key] = "done"
|
||
except Exception as e:
|
||
logger.error(f"mlx-whisper 模型下载失败: {model_size}, {e}")
|
||
_downloading[key] = "failed"
|
||
|
||
|
||
@router.post("/transcriber_download")
|
||
def download_transcriber_model(data: ModelDownloadRequest, background_tasks: BackgroundTasks):
|
||
"""触发后台下载指定的 whisper 模型。"""
|
||
if data.model_size not in WHISPER_MODEL_SIZES:
|
||
return R.error(msg=f"不支持的模型大小: {data.model_size}")
|
||
|
||
if data.transcriber_type == "mlx-whisper":
|
||
if platform.system() != "Darwin":
|
||
return R.error(msg="MLX Whisper 仅支持 macOS")
|
||
key = f"mlx-{data.model_size}"
|
||
if _downloading.get(key) == "downloading":
|
||
return R.success(msg="模型正在下载中")
|
||
background_tasks.add_task(_do_download_mlx_whisper, data.model_size)
|
||
else:
|
||
if _downloading.get(data.model_size) == "downloading":
|
||
return R.success(msg="模型正在下载中")
|
||
background_tasks.add_task(_do_download_whisper, data.model_size)
|
||
|
||
return R.success(msg="模型下载已开始")
|
||
|
||
|
||
@router.get("/sys_health")
|
||
async def sys_health():
|
||
"""结构化健康状态——任何子项异常都不应让整个 endpoint 5xx。
|
||
|
||
每个字段：'ok' | 'missing' | 'error'。
|
||
前端 useCheckBackend 用 /sys_check 做存活判定（不依赖外部依赖），
|
||
/sys_health 用来在设置页区分「后端没起」vs「后端起了但 ffmpeg 缺」vs「DB 写不进去」等更细的状态。
|
||
"""
|
||
ffmpeg_status = "ok"
|
||
try:
|
||
ensure_ffmpeg_or_raise()
|
||
except Exception:
|
||
ffmpeg_status = "missing"
|
||
|
||
db_status = "ok"
|
||
try:
|
||
from app.db.engine import engine
|
||
from sqlalchemy import text
|
||
with engine.connect() as conn:
|
||
conn.execute(text("SELECT 1"))
|
||
except Exception:
|
||
db_status = "error"
|
||
|
||
# 当前转写器配置 + 模型是否已下载（用 model.bin 落盘判定，与 transcriber 加载逻辑一致）
|
||
whisper_info: dict = {"size": None, "type": None, "downloaded": False, "checked": False}
|
||
try:
|
||
cfg = transcriber_config_manager.get_config()
|
||
size = cfg["whisper_model_size"]
|
||
ttype = cfg["transcriber_type"]
|
||
whisper_info["size"] = size
|
||
whisper_info["type"] = ttype
|
||
# 只有本地引擎才有「下载」概念；groq / bcut / kuaishou 在线引擎跳过
|
||
if ttype == "fast-whisper":
|
||
whisper_info["downloaded"] = _check_whisper_model_exists(size, "whisper")
|
||
whisper_info["checked"] = True
|
||
elif ttype == "mlx-whisper":
|
||
whisper_info["downloaded"] = _check_mlx_whisper_model_exists(size)
|
||
whisper_info["checked"] = True
|
||
except Exception:
|
||
pass
|
||
|
||
return R.success(data={
|
||
"backend": "ok",
|
||
"ffmpeg": ffmpeg_status,
|
||
"db": db_status,
|
||
"whisper_model": whisper_info,
|
||
})
|
||
|
||
|
||
@router.get("/sys_check")
|
||
async def sys_check():
|
||
"""轻量存活判定：后端进程能响应这个 endpoint 就算「起来了」，不查外部依赖。
|
||
|
||
给桌面端 useCheckBackend / Tauri ready-probe 用。
|
||
"""
|
||
return R.success()
|
||
|
||
|
||
@router.get("/deploy_status")
|
||
async def deploy_status():
|
||
"""返回部署监控所需的所有状态信息。
|
||
|
||
所有子项都用 try 包起来——监控页本身不应该被任何一个子项打死。
|
||
特别是 torch：它只在 fast-whisper 路径用得到，用 Groq / 必剪 / 快手在线
|
||
引擎的轻量部署完全可以不装，那种情况这个 endpoint 不应该 500。
|
||
"""
|
||
import os
|
||
|
||
# CUDA 状态
|
||
try:
|
||
import torch
|
||
cuda_available = torch.cuda.is_available()
|
||
cuda_info = {
|
||
"available": cuda_available,
|
||
"torch_installed": True,
|
||
"version": torch.version.cuda if cuda_available else None,
|
||
"gpu_name": torch.cuda.get_device_name(0) if cuda_available else None,
|
||
}
|
||
except Exception:
|
||
cuda_info = {
|
||
"available": False,
|
||
"torch_installed": False,
|
||
"version": None,
|
||
"gpu_name": None,
|
||
}
|
||
|
||
# Whisper 模型 / 转写器配置 + 本地下载状态
|
||
try:
|
||
transcriber_cfg = transcriber_config_manager.get_config()
|
||
size = transcriber_cfg["whisper_model_size"]
|
||
ttype = transcriber_cfg["transcriber_type"]
|
||
if ttype == "fast-whisper":
|
||
downloaded = _check_whisper_model_exists(size, "whisper")
|
||
elif ttype == "mlx-whisper":
|
||
downloaded = _check_mlx_whisper_model_exists(size)
|
||
else:
|
||
downloaded = False # 在线引擎无下载概念
|
||
whisper_info = {
|
||
"model_size": size,
|
||
"transcriber_type": ttype,
|
||
"downloaded": downloaded,
|
||
}
|
||
except Exception:
|
||
whisper_info = {"model_size": None, "transcriber_type": None, "downloaded": False}
|
||
|
||
# FFmpeg 状态
|
||
try:
|
||
ensure_ffmpeg_or_raise()
|
||
ffmpeg_ok = True
|
||
except Exception:
|
||
ffmpeg_ok = False
|
||
|
||
return R.success(data={
|
||
"backend": {"status": "running", "port": int(os.getenv("BACKEND_PORT", 8483))},
|
||
"cuda": cuda_info,
|
||
"whisper": whisper_info,
|
||
"ffmpeg": {"available": ffmpeg_ok},
|
||
}) |