{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/scaling-and-evaluating-sparse-autoencoders","title":"Scaling and evaluating sparse autoencoders","arxiv_id":"2406.04093","date":"2024-06-06","proceeding":null,"authors":["Leo Gao","Tom Dupré La Tour","Henk Tillman","Gabriel Goh","Rajan Troll","Alec Radford","Ilya Sutskever","Jan Leike","Jeffrey Wu"],"abstract":"Sparse autoencoders provide a promising unsupervised approach for extracting interpretable features from a language model by reconstructing activations from a sparse bottleneck layer. Since language models learn many concepts, autoencoders need to be very large to recover all relevant features. However, studying the properties of autoencoder scaling is difficult due to the need to balance reconstruction and sparsity objectives and the presence of dead latents. We propose using k-sparse autoencoders [Makhzani and Frey, 2013] to directly control sparsity, simplifying tuning and improving the reconstruction-sparsity frontier. Additionally, we find modifications that result in few dead latents, even at the largest scales we tried. Using these techniques, we find clean scaling laws with respect to autoencoder size and sparsity. We also introduce several new metrics for evaluating feature quality based on the recovery of hypothesized features, the explainability of activation patterns, and the sparsity of downstream effects. These metrics all generally improve with autoencoder size. To demonstrate the scalability of our approach, we train a 16 million latent autoencoder on GPT-4 activations for 40 billion tokens. We release training code and autoencoders for open-source models, as well as a visualizer.","url_abs":"https://arxiv.org/abs/2406.04093v1","url_pdf":"https://arxiv.org/pdf/2406.04093v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"scaling-and-evaluating-sparse-autoencoders","repo_url":"https://github.com/openai/sparse_autoencoder","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null},{"paper_slug":"scaling-and-evaluating-sparse-autoencoders","repo_url":"https://github.com/eleutherai/sae","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"scaling-and-evaluating-sparse-autoencoders","repo_url":"https://github.com/eleutherai/sparsify","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"scaling-and-evaluating-sparse-autoencoders","repo_url":"https://github.com/jkminder/dictionary_learning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"scaling-and-evaluating-sparse-autoencoders","repo_url":"https://github.com/saprmarks/dictionary_learning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"language-modelling","task_name":"Language Modelling"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"gpt-4","method_name":"GPT-4"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2406.04093","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2406.04093"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/eleutherai/sparsify","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/saprmarks/dictionary_learning","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/jkminder/dictionary_learning","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/openai/sparse_autoencoder","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/eleutherai/sae","reach":{"status":"ok","spdx":"MIT"}}],"summary":{"ran_draft_wrong":2,"unverified":8},"by_repo_kind":{"official":{"samples":1,"ran":1,"repositories":1},"listed":{"samples":9,"ran":1,"repositories":2}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"b4af06c57a251295","entry":"LN","repo":"openai/sparse_autoencoder","repo_kind":"official","path":"sparse_autoencoder/model.py","file_url":"https://github.com/openai/sparse_autoencoder/blob/HEAD/sparse_autoencoder/model.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"b4af06c57a251295"}},{"code_sha256_prefix":"3ca099bf6afb9a6a","entry":"geometric_median","repo":"saprmarks/dictionary_learning","repo_kind":"listed","path":"dictionary_learning/trainers/top_k.py","file_url":"https://github.com/saprmarks/dictionary_learning/blob/HEAD/dictionary_learning/trainers/top_k.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":"invariant","behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"3ca099bf6afb9a6a"}},{"code_sha256_prefix":"94434469ea2ebc0d","entry":"assert_type","repo":"eleutherai/sparsify","repo_kind":"listed","path":"sparsify/utils.py","file_url":"https://github.com/eleutherai/sparsify/blob/HEAD/sparsify/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"94434469ea2ebc0d"}},{"code_sha256_prefix":"3fafa18bde72510f","entry":"chunk_and_tokenize","repo":"eleutherai/sparsify","repo_kind":"listed","path":"sparsify/data.py","file_url":"https://github.com/eleutherai/sparsify/blob/HEAD/sparsify/data.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"3fafa18bde72510f"}},{"code_sha256_prefix":"c839c072b645f3c8","entry":"fused_encoder","repo":"eleutherai/sparsify","repo_kind":"listed","path":"sparsify/fused_encoder.py","file_url":"https://github.com/eleutherai/sparsify/blob/HEAD/sparsify/fused_encoder.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"c839c072b645f3c8"}},{"code_sha256_prefix":"b44a86959a6f33aa","entry":"get_columns_all_equal","repo":"eleutherai/sparsify","repo_kind":"listed","path":"sparsify/data.py","file_url":"https://github.com/eleutherai/sparsify/blob/HEAD/sparsify/data.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"b44a86959a6f33aa"}},{"code_sha256_prefix":"72500992c94f5911","entry":"get_layer_list","repo":"eleutherai/sparsify","repo_kind":"listed","path":"sparsify/utils.py","file_url":"https://github.com/eleutherai/sparsify/blob/HEAD/sparsify/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"72500992c94f5911"}},{"code_sha256_prefix":"55dfcfdde405bda2","entry":"get_saes_by_layer_name","repo":"eleutherai/sparsify","repo_kind":"listed","path":"sparsify/trainer.py","file_url":"https://github.com/eleutherai/sparsify/blob/HEAD/sparsify/trainer.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"55dfcfdde405bda2"}},{"code_sha256_prefix":"1930bbdc2f82dc62","entry":"quintic_newtonschulz","repo":"eleutherai/sparsify","repo_kind":"listed","path":"sparsify/muon.py","file_url":"https://github.com/eleutherai/sparsify/blob/HEAD/sparsify/muon.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"1930bbdc2f82dc62"}},{"code_sha256_prefix":"4cd2443146d52623","entry":"resolve_widths","repo":"eleutherai/sparsify","repo_kind":"listed","path":"sparsify/utils.py","file_url":"https://github.com/eleutherai/sparsify/blob/HEAD/sparsify/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"4cd2443146d52623"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}