| # Pesi a 8 bit (lineari simmetrici per canale): 518 MB → 262 MB, coseno con l'originale ≥ 0,998. | |
| import coremltools as ct | |
| import coremltools.optimize.coreml as cto | |
| m = ct.models.MLModel("EG2fixed512.mlpackage") | |
| q8 = cto.linear_quantize_weights(m, cto.OptimizationConfig( | |
| global_config=cto.OpLinearQuantizerConfig(mode="linear_symmetric", granularity="per_channel"))) | |
| q8.save("EG2-512-int8.mlpackage") | |