docs: update cifar_brevitas_training accuracy using representative ca…

…libration set

use_case_examples/cifar_brevitas_training/README.md

@@ -69,7 +69,7 @@ Testing with different rounding_threshold_bits values can help you understand th
 python3 evaluate_torch_cml.py --rounding_threshold_bits 1 2 3 4 5 6 7 8
 ```
 
-Using rounding with 6 bits for all accumulators provides a significant speedup for FHE, with only a 1.3% loss in accuracy compared to the original model. More details can be found in the Accuracy and Performance section below.
+Using rounding with 6 bits for all accumulators provides a significant speedup for FHE, with only a 2.7% loss in accuracy compared to the original model. More details can be found in the Accuracy and Performance section below.
 
 ## Fully Homomorphic Encryption (FHE)
 
@@ -93,11 +93,10 @@ Experiments were conducted on an m6i.metal machine offering 128 CPU cores and 51
 | ---------------------- | -------- | -------- |
 | VGG Torch              | None     | 88.7     |
 | VGG FHE (simulation\*) | None     | 88.7     |
-| VGG FHE (simulation\*) | 8 bits   | 88.3     |
-| VGG FHE (simulation\*) | 7 bits   | 88.3     |
-| VGG FHE (simulation\*) | 6 bits   | 87.5     |
-| VGG FHE (simulation\*) | 5 bits   | 84.9     |
-| VGG FHE                | 6 bits   | 87.5\*\* |
+| VGG FHE (simulation\*) | 8 bits   | 88.0     |
+| VGG FHE (simulation\*) | 7 bits   | 87.2     |
+| VGG FHE (simulation\*) | 6 bits   | 86.0     |
+| VGG FHE                | 6 bits   | 86.0\*\* |
 
 We ran the FHE inference over 10 examples and achieved 100% similar predictions between the simulation and FHE. The overall accuracy for the entire data-set is expected to match the simulation. The original model with a maximum of 13 bits of precision ran in around 9 hours on the specified hardware. Using the rounding approach, the final model ran in **31 minutes**, providing a speedup factor of 18x while preserving accuracy. This significant performance improvement demonstrates the benefits of the rounding operator in the FHE setting.
 

use_case_examples/cifar_brevitas_training/evaluate_torch_cml.py

@@ -1,5 +1,5 @@
 import argparse
-import pathlib
+from pathlib import Path
 
 import numpy as np
 import torch
@@ -11,11 +11,13 @@
 
 from concrete.ml.torch.compile import compile_brevitas_qat_model
 
+CURRENT_DIR = Path(__file__).resolve().parent
+
 
 def evaluate(torch_model, cml_model, device, num_workers):
 
     # Import and load the CIFAR test dataset (following bnn_pynq_train.py)
-    test_set = get_test_set(dataset="CIFAR10", datadir=".datasets/")
+    test_set = get_test_set(dataset="CIFAR10", datadir=CURRENT_DIR / ".datasets/")
     test_loader = DataLoader(test_set, batch_size=128, shuffle=False, num_workers=num_workers)
 
     torch_top_1_batches = []
@@ -76,7 +78,7 @@ def main(args):
     print("Device in use:", device)
 
     # Find relative path to this file
-    dir_path = pathlib.Path(__file__).parent.absolute()
+    dir_path = Path(__file__).parent.absolute()
 
     # Load checkpoint
     checkpoint = torch.load(
@@ -88,7 +90,7 @@ def main(args):
     model.load_state_dict(checkpoint["state_dict"], strict=False)
 
     # Load the training set
-    train_set = get_train_set(dataset="CIFAR10", datadir=".datasets/")
+    train_set = get_train_set(dataset="CIFAR10", datadir=CURRENT_DIR / ".datasets/")
 
     # Create a representative input-set from the training set that will be used used for both
     # computing quantization parameters and compiling the model