spandrel-extra-arches

0.2.0 · active · verified Thu Apr 16

spandrel-extra-arches is a Python library that implements various PyTorch model architectures with restrictive licenses, designed to extend the capabilities of the core `spandrel` package. It enables `spandrel` to automatically detect and load models that might have non-commercial or other specific license requirements, segregating them from `spandrel`'s permissively licensed architectures. The current version is 0.2.0, released on September 15, 2024, and it is actively maintained and developed in conjunction with `spandrel`.

Common errors

```
ModuleNotFoundError: No module named 'spandrel_extra_arches.architectures.<some_arch>'
```
cause The `spandrel-extra-arches` package is either not installed, or the specified architecture path is incorrect or does not exist. This can happen if attempting a direct import before installation or if an unsupported architecture name is used.

fix First, ensure `spandrel-extra-arches` is installed via `pip install spandrel-extra-arches`. If directly importing, double-check the exact import path and architecture name against the library's source code or documentation.
```
TypeError: <ArchitectureClass>() takes no positional arguments
```
cause This error occurs when attempting to instantiate an architecture class with positional arguments. A recent update in `spandrel` and `spandrel-extra-arches` requires all architecture parameters to be passed as keyword arguments.

fix Modify the instantiation call to use keyword arguments for all parameters, e.g., change `RCAN(2, 64, 16)` to `RCAN(scale=2, num_feat=64, num_rcab=16)`.

Warnings

gotcha Many architectures within `spandrel-extra-arches` are subject to restrictive licenses (e.g., non-commercial use only). Users engaging in commercial or closed-source projects MUST carefully review the individual licenses of each architecture they intend to use.
Fix: Always check the license of specific architectures before use, especially for commercial or closed-source applications.
breaking Starting with `spandrel` v0.4.0 (released alongside `spandrel-extra-arches` v0.2.0), all architectures now require keyword arguments for instantiation. Code that previously used positional arguments for initializing models might break.
Fix: Update model instantiation calls to use keyword arguments exclusively, e.g., `MyArch(scale=2, channels=3)` instead of `MyArch(2, 3)`.
gotcha While `spandrel-extra-arches` itself specifies `requires_python: >=3.8`, the core `spandrel` library, which is a dependency and the primary interface, has recently dropped support for Python 3.8 and 3.9, now requiring Python >=3.10. Using `spandrel-extra-arches` with older Python versions (3.8, 3.9) might lead to incompatibility issues with `spandrel`.
Fix: Ensure your Python environment is version 3.10 or newer when using `spandrel-extra-arches` with the latest `spandrel` package.

Install

pip install spandrel-extra-arches Install stable version

Imports

RCAN
```
from spandrel_extra_arches.architectures.rcan.__arch.rcan import RCAN
```
While direct import is possible for specific architectures, typical usage involves `spandrel`'s auto-detection mechanism after installation. This example demonstrates a direct architectural class import if needed.

Quickstart

This quickstart demonstrates how to use the `spandrel` library to load a model file, which would implicitly utilize the architectures provided by `spandrel-extra-arches` if one of its models is detected. The `spandrel` library handles the automatic detection and instantiation of the correct architecture.

import torch
from spandrel import Spandrel

# Ensure spandrel-extra-arches is installed (pip install spandrel-extra-arches)
# spandrel will automatically discover architectures provided by spandrel-extra-arches

# Example: Assuming you have a .pth model file for an architecture in spandrel-extra-arches
# For demonstration, let's pretend 'my_extra_arch_model.pth' is an RCAN model.
# In a real scenario, `spandrel.Spandrel()` would detect the architecture type.

try:
    # This path would be to an actual model file
    model_path = 'my_extra_arch_model.pth'
    # Create a dummy model file for demonstration if it doesn't exist
    if not torch.exists(model_path):
        # This is a placeholder. A real model would be loaded from a file.
        # In practice, you'd download a .pth file for an extra arch model.
        print(f"[INFO] No model file found at {model_path}. Skipping model loading example.")
        print("[INFO] Please provide a valid .pth model file for an architecture supported by spandrel-extra-arches to run this part.")
    else:
        # Load the model using spandrel
        spandrel_model = Spandrel(model_path)
        print(f"Model loaded successfully: {spandrel_model.name} - {spandrel_model.arch_name}")
        
        # Example of getting the underlying PyTorch module
        pytorch_module = spandrel_model.model
        print(f"PyTorch module type: {type(pytorch_module)}")

except Exception as e:
    print(f"An error occurred during model loading: {e}")

view raw JSON →