Cupy打包比预期中复杂，可能还涉及到dll其他的一些文件。多做点测试才能放出来
zelefans 发表于 2026-5-14 08:01
Cupy打包比预期中复杂，可能还涉及到dll其他的一些文件。多做点测试才能放出来 ...
我电脑也报错，为什么要找bin\\bin?
=== PTServer Diagnostic Report ===
generated: 2026-05-14 08:08:59
app_version: v0.1.0-alpha.1
ui_language: zh_CN
host: WIN-QNVME2PTKS9
os: Windows-11-10.0.26200-SP0
python: 3.12.12
executable: C:\VR Video Passthrough Server\VR_Video_Passthrough_Server\VR_Video_Passthrough_Server.exe
frozen: True
cwd: C:\VR Video Passthrough Server\VR_Video_Passthrough_Server
--- GPU (nvidia-smi) ---
nvidia-smi: Command '['nvidia-smi', '--query-gpu=name,driver_version,memory.total,memory.free,compute_cap', '--format=csv,noheader']' timed out after 5 seconds
--- Runtime libraries ---
onnxruntime: 1.25.1
ort_providers: TensorrtExecutionProvider,CUDAExecutionProvider,CPUExecutionProvider
cupy: (not installed: 系统找不到指定的文件。: 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v13.2\\bin\\bin')
cupy_devices: (unavailable: 系统找不到指定的文件。: 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v13.2\\bin\\bin')
numpy: 2.0.2
--- FFmpeg / FFprobe (PATH lookup) ---
ffmpeg: .\ffmpeg.EXE
ffmpeg version 2025-11-10-git-133a0bcb13-full_build-www.gyan.dev Copyright (c) 2000-2025 the FFmpeg developers
ffprobe: .\ffprobe.EXE
ffprobe version 2025-11-10-git-133a0bcb13-full_build-www.gyan.dev Copyright (c) 2007-2025 the FFmpeg developers
nvenc_encoders: h264_nvenc, hevc_nvenc
--- Warmup marker ---
path: C:\VR Video Passthrough Server\VR_Video_Passthrough_Server\runtime_cache\gpu_warmup_marker.json
(no warmup marker — first run not completed)
--- Last startup status (from /status) ---
phase : failed
step : failed
step_index : 1
step_total : 4
progress : 0.0
eta_sec : 45.0
elapsed_sec : 0.0
cold : True
is_known_slow : False
gpu_name :
compute_capability :
driver_version : 595.79
onnxruntime_version : 1.25.1
reason : inspect_failed
message : startup GPU warmup failed: 系统找不到指定的文件。: 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v13.2\\bin\\bin'
detail : 系统找不到指定的文件。: 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v13.2\\bin\\bin'
uptime_sec : 0.095
age_sec : 0.032
--- Recent server.log (last 200 lines) ---
path: C:\VR Video Passthrough Server\VR_Video_Passthrough_Server\debug_output\server.log
2026-05-14 08:06:58,072 startup_status: startup status listening on http://127.0.0.1:8299/status
2026-05-14 08:06:58,072 main: LAN_IP=192.168.71.233 HTTP_PORT=8200 UUID=2b543911-bd3f-5763-9764-01794e5fcd49
2026-05-14 08:06:58,072 main: VIDEO_DIRS=C:\VR Video Passthrough Server\VR_Video_Passthrough_Server\videos
2026-05-14 08:06:58,072 main: MODEL_PATH=C:\VR Video Passthrough Server\VR_Video_Passthrough_Server\models\rvm_mobilenetv3_fp32.onnx (exists=True)
2026-05-14 08:06:58,072 main: GPU_RUNTIME_CACHE=GpuRuntimeCacheEnv(runtime_cache_dir='C:\\VR Video Passthrough Server\\VR_Video_Passthrough_Server\\runtime_cache', cuda_cache_path='C:\\VR Video Passthrough Server\\VR_Video_Passthrough_Server\\runtime_cache\\cuda_compute_cache', cupy_cache_dir='C:\\VR Video Passthrough Server\\VR_Video_Passthrough_Server\\runtime_cache\\cupy', ort_cache_dir='C:\\VR Video Passthrough Server\\VR_Video_Passthrough_Server\\runtime_cache\\ort', marker_path='C:\\VR Video Passthrough Server\\VR_Video_Passthrough_Server\\runtime_cache\\gpu_warmup_marker.json')
2026-05-14 08:06:58,120 main: warmup prediction: cold=True reason=inspect_failed known_slow=False eta=45.0s gpu= cc= ort=1.25.1
2026-05-14 08:06:58,120 main: startup GPU warmup begin: force=False timeout=300.0s runs_per_shape=3
2026-05-14 08:06:58,120 main: startup GPU warmup failed; server will not start: 系统找不到指定的文件。: 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v13.2\\bin\\bin'
Traceback (most recent call last):
File "main.py", line 159, in main
File "utils\gpu_runtime_cache.py", line 458, in warmup_gpu_runtime_cache
File "utils\gpu_runtime_cache.py", line 224, in build_warmup_key
File "<frozen importlib._bootstrap>", line 1360, in _find_and_load
File "<frozen importlib._bootstrap>", line 1331, in _find_and_load_unlocked
File "<frozen importlib._bootstrap>", line 935, in _load_unlocked
File "pyimod02_importers.py", line 457, in exec_module
File "cupy\__init__.py", line 13, in <module>
_environment._setup_win32_dll_directory()# NOQA
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "cupy\_environment.py", line 294, in _setup_win32_dll_directory
os.add_dll_directory(cuda_bin_path)
File "<frozen os>", line 1154, in add_dll_directory
FileNotFoundError: 系统找不到指定的文件。: 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v13.2\\bin\\bin'
=== End of report ===
commade17402 发表于 2026-5-14 08:11
我电脑也报错，为什么要找bin\\bin?
因为cupy的原因，本项目关键性的优化需要，冷启动需要调用那个bin目录里的做简单编译。刚试了在项目里加个空的bin目录可以跳过它的检测，但无法处理。白天就重点检查了，我本机已经屏蔽了cuda变量，可以复现问题了
高效大佬必须点赞！坐等cuda整合包发布
大佬牛逼
VR连接服务器成功，但是打开视频显示一个全绿色的画面，一直转圈，电脑GPU利用率为0，4XVR
2026-05-14 11:48:14,798 pynv_stream: slate audio cache build failed interrupted=False rc=4294967274 elapsed=0.27s stderr= adts muxer supports only codec aac for type audio
Could not write header (incorrect codec parameters ?): Invalid argument
说是音频缓存构建失败
本帖最后由 zelefans 于 2026-5-14 12:52 编辑
joe123 发表于 2026-5-14 11:54
2026-05-14 11:48:14,798 pynv_stream: slate audio cache build failed interrupted= ...
这个问题应该是你看的视频里音频源不是AAC的。我当时扫描了硬盘里1000多个VR视频都是AAC音源所以直接用copy方法效率最高。这样的话要改音频的处理逻辑，有点把老的改坏的风险，你能否给一下你用的视频的下载位置。方便我做测试。 你再重启服务试试其他视频。
已扫描 F:\VR：Scanned videos: 1473Non-AAC audio videos: 0
目录下当前没有找到第一条音频流不是 AAC 的视频。我硬盘里真没……
zelefans 发表于 2026-5-14 12:10
这个问题应该是你看的视频里音频源不是AAC的。我当时扫描了硬盘里1000多个VR视频都是AAC音源所以直接用cop ...
我试了几个视频都是报这个错误，离线生成没问题，实时不行，应该不是视频的问题。
joe123 发表于 2026-5-14 13:19
我试了几个视频都是报这个错误，离线生成没问题，实时不行，应该不是视频的问题。 ...
离线那个音频最后是复制，那个不会出问题的。
我现在暂时没条件播放实时的，我先想办法能否模拟，不行下班再测。今天的升级按理不会触碰到音频的逻辑……
你看下runtime_cache\audio目录里有没有音频文件的缓存，文件大小是否正确，还说是空的没缓存成功一个
