GulkoA/TinyStories-tokenized-Llama-3.2

TinyStories dataset tokenized with Llama-3.2 Useful for accelerated training and testing of sparse autoencoders Context window: 128, not shuffled For first layer activations cache with Llama-3.2-1B, see GulkoA/TinyStories-Llama-3.2-1B-cache