Cached activations at layer 5 for gpt2 using dataset apollo-research/roneneldan-TinyStories-tokenizer-gpt2
Useful for accelerated training and testing of sparse autoencoders
context_window: 512 tokens
total_tokens: 51,200,000
batch_size: 8 prompts (4096 tokens)
layer_hook_name: blocks.5.hook_mlp_out
