Course 07

Accelerate Your Model

A functional model is useless if too slow or expensive to run. This course focuses on MLOps and optimization, particularly relevant for the African context where compute resources may be constrained and mobile deployment is essential.

Key Topics

  • Inference Optimization: Reducing latency
  • Quantization: Reducing model precision (8-bit, 4-bit)
  • Distillation: Training smaller student models
  • Hardware acceleration (TPUs/GPUs)
  • Mobile and edge deployment

Learning Outcomes

  • Understand the engineering challenges of deployment
  • Learn techniques to make models efficient for real-world use
  • Bridge the gap between research prototype and production
  • Deploy models on resource-constrained devices
  • Benchmark accuracy vs speed trade-offs

Practical Projects

  • Quantize models to 8-bit and benchmark accuracy loss
  • Knowledge Distillation: Teacher to Student model transfer
  • Deploy a quantized ASR model on mobile devices

Recommended Datasets

6 datasets are aligned with this course for hands-on exercises and projects.

Malaria Cell Images

Parasitized vs Uninfected Blood Cell Images (NIH)

image-classification computer-vision 27,558 images (~339MB)
English (Metadata)

KenSwQuAD

Swahili Question Answering Dataset

question-answering 7,526 QA pairs
Swahili

SIB-200

Benchmark for 200+ Languages

topic-classification benchmarking Based on FLORES-200
200+ including many African LRLs

AfriSpeech-200

Pan-African Speech Corpus

asr speech-recognition 200 hours
English (120 African accents)

IrokoBench

Comprehensive Benchmark for African Languages

benchmarking evaluation Multiple evaluation sets
Hausa Igbo Yoruba Swahili +2 more

Nigerian Colposcopy Images

Cervical Cancer Screening Images from Nigeria

image-classification computer-aided-diagnosis 3,356 images
English (Metadata)

About This Course

Many deployment scenarios in Africa involve limited computational resources. This course teaches you to make models smaller, faster, and more efficient without sacrificing too much performance.

Key Topics

  • Quantization: Reducing model precision for efficiency
  • Pruning: Removing unnecessary model parameters
  • Knowledge Distillation: Training smaller models from larger ones
  • Efficient Architectures: MobileNet, DistilBERT, and similar models
  • Edge Deployment: Running models on mobile and IoT devices

Prerequisites

  • Completion of Courses 01-06 or equivalent knowledge
  • Understanding of model architectures
  • Experience with deployment tools