SeedLM: Compressing LLM Weights into Seeds of Pseudo-Random Generators

This sound like compression with extra steps.. What makes this technique particular to LLM weights instead of general purpose data?

Weights in neural networks don't always need to be precise. Not all weights are equally useful to the network. There seems to be a lot of redundancy that can be replaced with approximations.

This technique seems a bit similar to lossy image compression that replaces exact pixels with a combination of pre-defined patterns (DCT in JPEG), but here the patterns aren't from cosine function, but from a pseudo-random one.

It may also be beating simple quantization from just adding noise that acts as dithering, and breaks up the bands created by combinations of quantized numbers.

> What makes this technique particular to LLM weights

This is my understanding as a non-expert.

LLM activations tend to be relatively sparse with large outliers. With linear quantization, this means you either have to clip off the outliers or you have to stretch your range to include the outliers, which wastes precious bits. Neither of these works well, so essentially all LLM quantization research is using various heuristics to get around these outliers. For example, you can do linear quantization but split the activations up into smaller blocks to make it less likely that any given block contains an outlier.

Another trick people have discovered (predates LLMs) is applying a random rotation/projection to the embeddings. This has the effect of making sure no one dimension in the vector dominates the others (which again hurts quantization). This works because in order for a single dimension to dominate, all the others have to "conspire" to be near zero. When you have 10,000+ dimensions, that's very unlikely.

This paper applies the latter trick. Instead of pre-generating the random projection matrices, they generate them on the fly on the accelerator from a seed that is fixed for each block. The seed is chosen from an offline brute-force search that needs only the weights of the network. This separates it from a lot of other quantization methods that either require calibration data or have to be simulated at training time so the network learns the quantization parameters itself.

You might think this is wasteful/might hurt performance, but it turns out that LLM inference is heavily memory-bound as it involves streaming a very large neural network into the accelerator (GPU/TPU/NPU/whatever) to operate on a relatively small amount of data, so there are lots of "free cycles" to generate these random numbers. Of course, if you care about power usage that might not be a great idea...