Compression Method¶
netspresso.enums.CompressionMethod
¶
Bases: StrEnum
Available Compression Methods¶
| Name | Value | Description | Type |
|---|---|---|---|
| PR_L2 | PR_L2 | L2 Norm Pruning | Pruning |
| PR_GM | PR_GM | GM Pruning | Pruning |
| PR_NN | PR_NN | Nuclear Norm Pruning | Pruning |
| PR_SNP | PR_SNP | Structured Neuron-level Pruning | Pruning |
| PR_ID | PR_ID | Pruning By Index | Pruning |
| FD_TK | FD_TK | Tucker Decomposition | Factorization |
| FD_SVD | FD_SVD | Singular Value Decomposition | Factorization |
| FD_CP | FD_CP | CP Decomposition | Factorization |
Framework Compatibility¶
Warning
- Nuclear Norm (PR_NN) is only supported in the Tensorflow-Keras framework.
- Structured Neuron-level (PR_SNP) is only supported in the PyTorch and ONNX frameworks.
- All other methods are supported across all frameworks.
Example¶
from netspresso.enums import CompressionMethod
# Select a compression method
compression_method = CompressionMethod.PR_L2
# Available methods
print(f"L2 Norm Pruning: {CompressionMethod.PR_L2}")
print(f"GM Pruning: {CompressionMethod.PR_GM}")
print(f"Tucker Decomposition: {CompressionMethod.FD_TK}")
print(f"CP Decomposition: {CompressionMethod.FD_CP}")