CP 2.9.5 + Nvidia 50xx - Install Pytorch nightly?

T

TSAN

Young grasshopper
Jul 16, 2019
33
8
USA
Recent PC build with Linux mint. Initially got CP 2.9.5 / Yolo 5.6.2 working with old 1070 ti.
(For anyone on Linux mint struggling - the modules get installed with path bug. The modules / runtime get installed with path "linuxmint" no space but modules are looking for "linux mint" with space. Adding space to directories fixes.
So had it working for ~week on 1070 ti

But now upgraded to 5070 ti and new Yolo 5.6.2 issues.

16:27:53:detect_adapter.py: /usr/bin/codeproject.ai-server-2.9.5/runtimes/bin/linux mint/python38/venv/lib/python3.8/site-packages/torch/cuda/__init__.py:155: UserWarning:
16:27:53:detect_adapter.py: NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
16:27:53:detect_adapter.py: The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70 sm_75 sm_80 sm_86.
16:27:53:detect_adapter.py: If you want to use the NVIDIA GeForce RTX 5070 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
16:27:53:detect_adapter.py: warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))
16:27:53:detect_adapter.py: YOLOv5.1m summary: 391 layers, 21805053 parameters, 0 gradients
16:27:53:detect_adapter.py: Adding AutoShape...

I'm not clear on instructions at - Start Locally

But further review I need pytorch nightly build. Not the stable.

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch


Conflicting python versions?
Your assumption I have no idea what I'm doing with pip + pytorch is 100% accurate as I have not spent time learning the nuts and bolts of CP.
Is there a way to amend Yolo 5.6.2 module setup script so when it sources pytorch it download/installs the pytorch nightly/cu128 build?

Also. Not been successful with Yolov8 on 5070 ti either (But I never used v8 in the past either.)
It does run without error. However BI returns AI error 500 and within 5 minutes Yolov 8 crashes.

OS: Linux Mint 22.1 x86_64
Kernel: 6.11.0-21-generic
AMD Ryzen 9 9950X3D (32) @ 5.752GHz
Memory: 44508MiB / 94171MiB
NVIDIA GeForce RTX 5070 Ti
NVIDIA-SMI 570.124.06
CUDA Version: 12.8

Thank you.
 
TSAN said:
Recent PC build with Linux mint. Initially got CP 2.9.5 / Yolo 5.6.2 working with old 1070 ti.
(For anyone on Linux mint struggling - the modules get installed with path bug. The modules / runtime get installed with path "linuxmint" no space but modules are looking for "linux mint" with space. Adding space to directories fixes.
So had it working for ~week on 1070 ti

But now upgraded to 5070 ti and new Yolo 5.6.2 issues.

16:27:53:detect_adapter.py: /usr/bin/codeproject.ai-server-2.9.5/runtimes/bin/linux mint/python38/venv/lib/python3.8/site-packages/torch/cuda/__init__.py:155: UserWarning:
16:27:53:detect_adapter.py: NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
16:27:53:detect_adapter.py: The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70 sm_75 sm_80 sm_86.
16:27:53:detect_adapter.py: If you want to use the NVIDIA GeForce RTX 5070 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
16:27:53:detect_adapter.py: warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))
16:27:53:detect_adapter.py: YOLOv5.1m summary: 391 layers, 21805053 parameters, 0 gradients
16:27:53:detect_adapter.py: Adding AutoShape...

I'm not clear on instructions at - Start Locally

But further review I need pytorch nightly build. Not the stable.

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch


Conflicting python versions?
Your assumption I have no idea what I'm doing with pip + pytorch is 100% accurate as I have not spent time learning the nuts and bolts of CP.
Is there a way to amend Yolo 5.6.2 module setup script so when it sources pytorch it download/installs the pytorch nightly/cu128 build?

Also. Not been successful with Yolov8 on 5070 ti either (But I never used v8 in the past either.)
It does run without error. However BI returns AI error 500 and within 5 minutes Yolov 8 crashes.

OS: Linux Mint 22.1 x86_64
Kernel: 6.11.0-21-generic
AMD Ryzen 9 9950X3D (32) @ 5.752GHz
Memory: 44508MiB / 94171MiB
NVIDIA GeForce RTX 5070 Ti
NVIDIA-SMI 570.124.06
CUDA Version: 12.8

Thank you.
Click to expand...
Just use the Object Detection (YOLOv5 .NET) module it will use your GPU and should be faster then Object Detection (YOLOv5 6.2)
 
MikeLud1 said:
Just use the Object Detection (YOLOv5 .NET) module it will use your GPU and should be faster then Object Detection (YOLOv5 6.2)
Click to expand...

Yes Yolo5.NET does launch CUDA enabled and does work with BI. Hmm - I'll have to learn about this as most my experience with BI is Yolo 5.6.2 with ipcam-combined model with specific object confirmations within BI.
But this use case is simple for on demand run when leave home for few indoor cameras to detect person its good to go.

Did something change with Yolo5.NET to work CUDA? I have another 8 cam BI instance I support with Server22 + A2000 and use Yolo 5.6.2. On that system YOLOv5.NET in past would only run CPU which is no go on that system because its on older ESXi server with several VMs. Needs A2000 offload.
I did just jump many CP versions (2.6.2? to 2.9.5) on it a few weeks ago on it because of CP updating / offline server issue. Update was postponed. Don't believe I tried running YOLOv5 on that system since 2.9.5 update

Interesting. Faster eh? Showing 20-36ms in log presently.
The description of YOLO5.NET points that its the go to for CPU and not CUDA " This module is best for those on Windows and Linux without CUDA enabled GPUs"
But Yolov5.NET is also more ideal for than 5.6.2 or v8 for CUDA enabled systems?

Thanks.
 
You must log in or register to reply here.
Top Bottom