BatchTopK Sparse Autoencoders
This is an incremental improvement for researchers and practitioners using sparse autoencoders to interpret language models, offering better reconstruction with tunable sparsity.
The paper tackles the problem of improving sparse autoencoders for interpreting language model activations by introducing BatchTopK SAEs, which relax the top-k constraint to batch-level to allow variable active latents per sample. This method consistently outperforms TopK SAEs in reconstructing activations from GPT-2 Small and Gemma 2 2B, achieving comparable performance to state-of-the-art JumpReLU SAEs while enabling direct specification of average latents.
Sparse autoencoders (SAEs) have emerged as a powerful tool for interpreting language model activations by decomposing them into sparse, interpretable features. A popular approach is the TopK SAE, that uses a fixed number of the most active latents per sample to reconstruct the model activations. We introduce BatchTopK SAEs, a training method that improves upon TopK SAEs by relaxing the top-k constraint to the batch-level, allowing for a variable number of latents to be active per sample. As a result, BatchTopK adaptively allocates more or fewer latents depending on the sample, improving reconstruction without sacrificing average sparsity. We show that BatchTopK SAEs consistently outperform TopK SAEs in reconstructing activations from GPT-2 Small and Gemma 2 2B, and achieve comparable performance to state-of-the-art JumpReLU SAEs. However, an advantage of BatchTopK is that the average number of latents can be directly specified, rather than approximately tuned through a costly hyperparameter sweep. We provide code for training and evaluating BatchTopK SAEs at https://github.com/bartbussmann/BatchTopK