DEV Community

From code to weights Series' Articles

Back to Carlos Chinchilla Corbacho's Series
A model doesn't read text: what a tokenizer decides for you
Cover image for A model doesn't read text: what a tokenizer decides for you

A model doesn't read text: what a tokenizer decides for you

2
Comments 7
8 min read
An embedding is a lookup table, and everything else is how you fill it
Cover image for An embedding is a lookup table, and everything else is how you fill it

An embedding is a lookup table, and everything else is how you fill it

1
Comments 4
9 min read
Shapes are the type system of ML, and nobody checks them for you
Cover image for Shapes are the type system of ML, and nobody checks them for you

Shapes are the type system of ML, and nobody checks them for you

Comments 2
10 min read
Attention is a learned weighted average, and the cost is in the square
Cover image for Attention is a learned weighted average, and the cost is in the square

Attention is a learned weighted average, and the cost is in the square

Comments
10 min read
A transformer block is six tensors and a bus: read it like a pipeline
Cover image for A transformer block is six tensors and a bus: read it like a pipeline

A transformer block is six tensors and a bus: read it like a pipeline

Comments
10 min read
GPT-2 is the transformer plus five decisions, and inference is a different program
Cover image for GPT-2 is the transformer plus five decisions, and inference is a different program

GPT-2 is the transformer plus five decisions, and inference is a different program

Comments 5
9 min read
From GPT-2 to Llama: four changes and the reason for each
Cover image for From GPT-2 to Llama: four changes and the reason for each

From GPT-2 to Llama: four changes and the reason for each

Comments
9 min read