DeepCompressionStats
Models & Types · AiDotNet.ModelCompression
Statistics about Deep Compression performance.
For Beginners
These statistics help you understand how well compression worked:
- CompressionRatio: How much smaller the model is (e.g., 35x means 35 times smaller)
- Sparsity: What fraction of weights are zero (e.g., 0.9 = 90% zeros)
- BitsPerWeight: How many bits are used per non-zero weight
Constructors
| Constructor | Summary |
|---|---|
DeepCompressionStats() |
Properties
| Property | Summary |
|---|---|
double BitsPerWeight { get; set; } |
Effective bits per weight after quantization. |
double CompressionRatio { get; set; } |
Overall compression ratio (original / compressed). |
double PruningRatio { get; } |
Compression ratio from pruning stage alone. |
double QuantizationRatio { get; } |
Compression ratio from quantization stage alone. |
double Sparsity { get; set; } |
Sparsity achieved by pruning (fraction of zeros). |
int NumClusters { get; set; } |
Number of clusters used for quantization. |
long CompressedSizeBytes { get; set; } |
Compressed size in bytes after all three stages. |
long OriginalSizeBytes { get; set; } |
Original size in bytes before compression. |