Datasetpublic

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

sourceHugging Facecdla-sharing-1.0updated 2y agoView on Hugging Face

1likes125downloads

Dataset Card

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