EnragedAntelopeandSisyphus
8573bed33a
Bump version to 0.6.0 for standalone loading release. Update requirements for ModelOpt 0.43.0 compatibility. Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
48 lines
1.4 KiB
Python
48 lines
1.4 KiB
Python
"""
|
|
ComfyUI NVIDIA ModelOpt Integration
|
|
|
|
Custom nodes for quantizing and loading models optimized with NVIDIA ModelOpt.
|
|
|
|
Installation:
|
|
1. Place this folder in ComfyUI/custom_nodes/
|
|
2. Install dependencies: pip install -r requirements.txt
|
|
3. Restart ComfyUI
|
|
|
|
For more information, see: https://github.com/EnragedAntelope/comfy-modelopt
|
|
"""
|
|
|
|
import warnings
|
|
import os
|
|
import sys
|
|
|
|
# Suppress misleading ModelOpt FP8 CUDA extension build warning on Windows.
|
|
# ModelOpt tries to JIT-compile a CUDA extension for FP8 using system nvcc.
|
|
# This fails on Windows with modern MSVC (unsupported compiler version),
|
|
# but it gracefully falls back to simulated/fake FP8 which works correctly.
|
|
# The warning is multi-line and noisy, so we suppress it by module.
|
|
warnings.filterwarnings(
|
|
"ignore",
|
|
category=UserWarning,
|
|
module="modelopt.torch.utils.cpp_extension",
|
|
)
|
|
|
|
# Also suppress the distributed/elastic multiprocessing redirect warning
|
|
# that appears on Windows (harmless, but clutters output)
|
|
warnings.filterwarnings(
|
|
"ignore",
|
|
message="Redirects are currently not supported in Windows or MacOs",
|
|
)
|
|
|
|
# Suppress setuptools distutils deprecation warning on Windows
|
|
warnings.filterwarnings(
|
|
"ignore",
|
|
message="_get_vc_env is private",
|
|
)
|
|
|
|
from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
|
|
|
|
__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]
|
|
|
|
# Display startup message
|
|
print("NVIDIA ModelOpt for ComfyUI loaded successfully")
|