Updating ComfyUI causes Torch re-install

[SiegeKeebsOffical]

Package

ComfyUI

When did the issue occur?

Updating the Package

What GPU / hardware type are you using?

ASUS STRIX 395+ Radeon 8060S

What happened?

I have to use a custom version of torch since I run AMD, every time I update ComfyUI I have to uninstall torch and reinstall my rocm torch version.

Is there a way around this?

Console output

When running ComfyUI I also get this:

WARNING: Potential Error in code: Torch already imported, torch should never be imported before this point.

Version

any version

What Operating System are you using?

Linux

[NeuralFault]

When you say "custom version" do you mean the nightly build from ROCm/TheRock releases? or custom compiled?

Currently, Stability Matrix on linux uses official PyTorch builds from upstream PyTorch (2.10-rocm7.1). Does this build cause issues with Strix Halo/gfx1151? Or is it getting overwritten to the CPU build of PyTorch (the default for linux officially)?

[jberube]

I have the same message at boot time of ComfyUI, but I'm on a Cuda stack.

Prestartup times for custom nodes:
   0.0 seconds: /home/test/doh/ai/apps/ComfyUI/custom_nodes/rgthree-comfy
   0.0 seconds: /home/test/doh/ai/apps/ComfyUI/custom_nodes/ComfyUI-DepthAnythingV3
   0.0 seconds: /home/test/doh/ai/apps/ComfyUI/custom_nodes/comfyui-easy-use
   0.9 seconds: /home/test/doh/ai/apps/ComfyUI/custom_nodes/ComfyUI-SAM3DBody
   1.2 seconds: /home/test/doh/ai/apps/ComfyUI/custom_nodes/ComfyUI-Manager

WARNING: Potential Error in code: Torch already imported, torch should never be imported before this point.
Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend cuda: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Checkpoint files will always be loaded safely.
Total VRAM 24072 MB, total RAM 128506 MB
pytorch version: 2.10.0+cu130
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 4090 : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 122080.0
Using sage attention
aimdo: src/control.c:68:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 4090 (VRAM: 24071 MB)
DynamicVRAM support detected and enabled
Python version: 3.12.11 (main, Aug 31 2025, 22:52:41) [GCC 11.4.0]
ComfyUI version: 0.16.4
ComfyUI frontend version: 1.39.19
[Prompt Server] web root: /home/test/doh/ai/apps/ComfyUI/.venv/lib/python3.12/site-packages/comfyui_frontend_package/static