Week #5 — Training the Model
Sachin C -
Welcome back! This week, I decided to begin finalizing my dataset and training my AI model. The very first thing I (and my father) have noticed is how much energy training the model consumes.
How Training Works
Training an AI model involves feeding it labeled images, allowing it to adjust its internal parameters (weights and biases) to recognize patterns over time. The training process consists of multiple iterations, known as epochs, where the model continuously improves its accuracy.
One thing I quickly realized is that AI training is incredibly energy-intensive. Even though I am training a lightweight model, running MobileNet on a high-performance GPU (RTX 4070 Super) can draw anywhere from 100 to 300 watts per hour, depending on the hardware. Thankfully, my household recently switched to solar energy, so I have been trying to train during bright and sunny days (although with the recent showers that has fallen slightly flat 🙂)
Keep in Mind:
Training large AI models like GPT-4 consumes millions of kilowatt-hours (kWh), comparable to the annual energy usage of thousands of homes.
A single training session for MobileNet on my dataset takes several hours and consumes roughly 2-5 kWh, depending on batch sizes and processing power.
This has made me even more aware of the importance of energy-efficient AI, and has given me brand-new motivation to continue my project. If we can minimize the amount of energy used in the usage stages, we can invest more energy into the training. This allows for us to sustainably advance in the tech sector.
Thank you for checking back in with me! I hope to see you next week.
Comments:
All viewpoints are welcome but profane, threatening, disrespectful, or harassing comments will not be tolerated and are subject to moderation up to, and including, full deletion.