| import numpy as np | |
| # SpTn-Model-only-1P | |
| # Capacity experiment | |
| print("================================") | |
| print("SpTn-Model-only-1P") | |
| print("Capacity Test") | |
| print("================================") | |
| # Exactly ONE parameter | |
| w = np.float32(0.5) | |
| values = np.linspace(0.0, 1.0, 1000001, dtype=np.float32) | |
| # Find how many distinct float32 values are representable | |
| unique_count = len(np.unique(values)) | |
| print("Trainable parameters:", 1) | |
| print("Parameter dtype:", w.dtype) | |
| print("Parameter size:", w.nbytes, "bytes") | |
| print("Distinct tested values:", unique_count) | |
| # Information capacity approximation | |
| bits = np.log2(unique_count) | |
| print("Approximate information capacity:") | |
| print(bits, "bits") | |
| print("================================") |