---
license: mit
datasets:
- uoft-cs/cifar10
language:
- en
pipeline_tag: image-classification
tags:
- efficient-inference
- computer-vision
- efficient-ml
- resnet
- cifar10
- pruning
- channel-gating
- neuromorphic
metrics:
- latency
- throughput
- peak-gpu-memory
- parameters
- accuracy
library_name: pytorch
base_model: microsoft/resnet-18
model-index:
- name: NeuroLattice ResNet-18
  results:
  - task:
      type: image-classification
    dataset:
      name: CIFAR-10
      type: cifar10
    metrics:
    - type: accuracy
      value: 91.24
      name: Top-1 Accuracy (%)
    - type: parameters-reduction-vs-baseline
      value: 78.8
      name: model-parameters-reduction (%)
    - type: speedup-vs-fp32-baseline
      value: 76.2
      name: Speedup vs fp32 Baseline (x)
    - type: memory-reduction-vs-fp32
      value: 83.4
      name: Memory Reduction vs fp32 Baseline (%)
    - type: latency
      value: 0.025
      name: inference-latency (ms/sample)
    - type: parameters 
      value: 2373455
      name: model-parameter
---
# Model Card — NeuroLattice™ ResNet-18

**NeuroLattice™ ResNet-18** is a production-optimized image classification model delivering substantial inference efficiency gains while maintaining high accuracy on CIFAR-10.  
The model is designed for **enterprise deployment**, prioritizing **low latency**, **high throughput**, and **minimal GPU memory usage** under real-world inference workloads.

The results demonstrate clear operational advantages over standard ResNet-18 baselines, validated under identical hardware and evaluation conditions.

## Performance Overview
![inference_comparison](https://cdn-uploads.huggingface.co/production/uploads/66af2c5b491b555fef86c068/AXHcUhj9H3ObUXiwf0V-R.png)

![output (9)](https://cdn-uploads.huggingface.co/production/uploads/66af2c5b491b555fef86c068/-Ek5NOX06Ns9eb_mpXy0O.png)
![output (8)](https://cdn-uploads.huggingface.co/production/uploads/66af2c5b491b555fef86c068/T559WHTiXQgdRXBMaQpaA.png)
![output (5)](https://cdn-uploads.huggingface.co/production/uploads/66af2c5b491b555fef86c068/VK3dU3PRO4GNtGN8qLhW5.png)
**Evaluation Context**
- Dataset: CIFAR-10  
- Input Resolution: 32 × 32  
- Batch Size: 4096  
- Samples Evaluated: 10,000  
- GPU: NVIDIA GeForce RTX 4050 (Laptop, 6 GB)


## How to Get Started with the Model

### Installation

```bash
pip install -r requirements.txt
```
#### Prerequisites

- **Python:** 3.8 or higher (tested with Python 3.12.7)
- **CUDA:** Optional, for GPU acceleration (CUDA 11.8+ recommended)

#### RUN
```bash
$env:KMP_DUPLICATE_LIB_OK="TRUE"; python hf_inference_resnet_standalone.py --checkpoint model.pt --batch-size 4096 --evaluate --plot
```

## Model Overview

- **Model Name:** NeuroLattice™ ResNet-18  
- **Task:** Image Classification  
- **Dataset:** CIFAR-10  
- **Accuracy:** 91.24%  
- **Inference Precision:** FP16  
- **License:** MIT  

This model belongs to the ResNet-18 family and is engineered for **deterministic, high-efficiency inference**.  
Design emphasis is placed on **scalability**, **resource efficiency**, and **predictable runtime performance**.

## Business Impact

NeuroLattice™ ResNet-18 enables organizations to:
- Reduce infrastructure and GPU memory costs
- Increase inference density per device
- Achieve lower latency without sacrificing accuracy
- Deploy deep learning models in constrained environments

The model is production-ready and designed for seamless integration into existing inference systems.