What's next? Next word prediction with PyTorch

Today, I want to take you through a simple next-word prediction model built using PyTorch. These models are famous in search applications, Gmail’s Smart Compose feature, which predicts the next words or sentences in emails you want to write. Google goes into detail on how they build their smart compose feature in their research blog post here. From this, I want to pull out some key learnings and requirements: Latency. Latency is important, must generate a response in under 100ms....

<span title='2023-08-10 00:13:04 +0000 UTC'>August 10, 2023</span>&nbsp;·&nbsp;5 min&nbsp;·&nbsp;James Malcolm

PyTorch Autograd and Training Models

Recently I’ve found myself wanting to dig a bit deeper into PyTorch to really understand how it works and hopefully figure out how to expand on my skillset to try new things. To aid this, I’ve been following the Deep Learning with PyTorch book by Eli Stevens, Luca Antiga and Thomas Viehmann. The book itself has been great and highly useful. I recommend it to anyone. In my observations, one the hardest things for newer users of PyTorch to understand is how to train a model....

<span title='2021-02-01 08:06:57 +0000 UTC'>February 1, 2021</span>&nbsp;·&nbsp;6 min&nbsp;·&nbsp;James Malcolm