ninovo/efficientnet-b1-pruned-int8-cpu

EfficientNet-B1 Pruned INT8 CPU-Only Baseline

This model is a CPU-optimized image classification baseline exported to ONNX for low-latency inference.

Backbone: EfficientNet-B1
Scope: full image classification model exported to ONNX
Optimization: structured pruning + dynamic INT8 quantization
Input size: 192 × 192

Parameters: 5.23M
Model size:

• FP32: 19.96 MB
• INT8: 5.45 MB

Labels

Depends on the classifier head used in training.
If exported from ImageNet-pretrained EfficientNet-B1, labels follow ImageNet.
If fine-tuned, labels follow the custom dataset.

Load

!pip install -U onnx onnxscript onnxruntime
from huggingface_hub import hf_hub_download
import onnxruntime as ort

model_path = hf_hub_download(
    repo_id="ninovo/efficientnet-b1-pruned-int8-cpu",
    filename="efficientnet_b1_192_pruned_int8_dynamic.onnx"
)

session = ort.InferenceSession(
    model_path,
    providers=["CPUExecutionProvider"]
)