Do you mean the number of tokens in the LLM’s tokenizer, or the dictionary size of the compression algorithm?
The vocab size of the pretrained models is not mentioned anywhere in the paper. Although, they did conduct an experiment where they measured compression performance while using tokenizers of different vocabulary sizes.
If you meant the dictionary size of the compression algorithm, then there was no dictionary because they only used arithmetic coding to do the compression which doesn’t use dictionaries.
Yes. They also mention that using such large models for compression is not practical because their size thwarts any amount of data you might want to compress. But this result gives a good picture into how generalized such large models are, and how well they are able to predict the next tokens for image/audio data at a high accuracy.
Interesting. I’m just thinking aloud to understand this.
In this case, the models are looking at a few sequence of bytes in their context and are able to predict the next byte(s) with good accuracy, which allows efficient encoding. Most of our memories are associative, i.e. we associate them with some concept/name/idea. So, do you mean, our brain uses the concept to predict a token which gets decoded in the form of a memory?
I don’t know. I have found that the folks on Technology community appreciate many of my computer science posts. But a dedicated Comp Science community which is active, will be awesome.
Hi @qwop, I am the author. Thank you for reading and the kind words. I would like to understand the error I made better so that I don’t repeat in future, and if I can fix it. Could you please clarify?
Thank you! That’s helpful. I spent quite some time trying to understand the difference between UTF-8 and Python’s representation and arrived at the same understanding as you wrote. However, most of the external documents simply say that strings in Python are UTF-8 which made me conclude that perhaps I am missing something and it might be safer to write it as utf-8.
How Python Compares Floats and Ints: When Equals Isn’t Really Equal (blog.codingconfessions.com)
A Deep Dive into the Underlying Architecture of Groq's LPU (codeconfessions.substack.com)
CPython Type System Internals: Video Series (codeconfessions.substack.com)
CPython Dynamic Dispatch Internals: What Happens When You do a + b? (codeconfessions.substack.com)
CPython Object System Internals: Understanding the Role of PyObject (codeconfessions.substack.com)
A Linear Algebra Trick for Computing Fibonacci Numbers Fast (codeconfessions.substack.com)
What Every Developer Should Know About GPU Computing (codeconfessions.substack.com)
A primer on GPU architecture and computing
An Analysis of DeepMind's 'Language Modeling Is Compression' Paper (codeconfessions.substack.com)
How to build a computer using origami (apieceofthepi.substack.com)
How CPython Implements and Uses Bloom Filters for String Processing (codeconfessions.substack.com)
Inside CPython’s Clever Use of Bloom Filters for Efficient String Processing