{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:BEFU2LW7MUBON53YBE2PPLF7EX","short_pith_number":"pith:BEFU2LW7","schema_version":"1.0","canonical_sha256":"090b4d2edf6502e6f7780934f7acbf25d1e69812428857dcfc13eef8921aa266","source":{"kind":"arxiv","id":"2606.26620","version":1},"attestation_state":"computed","paper":{"title":"Discovering Millions of Interpretable Features with Sparse Autoencoders","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bing Zhao, Hu Wei, Lin Qu, Wei Wang, Weixu Qiao, Wenbo Li, Xinyang He, Xuan Ren","submitted_at":"2026-06-25T05:33:03Z","abstract_excerpt":"Sparse autoencoders (SAEs) have emerged as a powerful tool for decomposing superposed language model representations into sparse and interpretable features. However, training SAEs is computationally expensive, and available open-source SAE models remain limited. In this work, we introduce \\textbf{Qwen3-Instruct SAE}, a comprehensive suite of SAEs trained on the Qwen3 instruction-tuned model family, covering Qwen3-1.7B, Qwen3-4B, and Qwen3-8B. For Qwen3-1.7B and Qwen3-4B, we train layer-wise SAEs at three key activation sites: residual streams, MLP outputs, and attention outputs. For Qwen3-8B, "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2606.26620","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-25T05:33:03Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c8532f31defb6a801ba4d2b737736277ae5d6022065d8f0a7599c2a0fac49549","abstract_canon_sha256":"a23063b7e160a156bc16326c4174aa456e6358b6fea28c92ecfb1aa9e44c9428"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T01:15:36.437886Z","signature_b64":"87diX7+y3oyqF/++6u98Y9sEP0jz3dqEBOodvgwWqvrfkyVjzDFH+Q4GNAtOMAcfgwnPljq+SyvpVob5Mx4HAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"090b4d2edf6502e6f7780934f7acbf25d1e69812428857dcfc13eef8921aa266","last_reissued_at":"2026-06-26T01:15:36.437508Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T01:15:36.437508Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Discovering Millions of Interpretable Features with Sparse Autoencoders","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bing Zhao, Hu Wei, Lin Qu, Wei Wang, Weixu Qiao, Wenbo Li, Xinyang He, Xuan Ren","submitted_at":"2026-06-25T05:33:03Z","abstract_excerpt":"Sparse autoencoders (SAEs) have emerged as a powerful tool for decomposing superposed language model representations into sparse and interpretable features. However, training SAEs is computationally expensive, and available open-source SAE models remain limited. In this work, we introduce \\textbf{Qwen3-Instruct SAE}, a comprehensive suite of SAEs trained on the Qwen3 instruction-tuned model family, covering Qwen3-1.7B, Qwen3-4B, and Qwen3-8B. For Qwen3-1.7B and Qwen3-4B, we train layer-wise SAEs at three key activation sites: residual streams, MLP outputs, and attention outputs. For Qwen3-8B, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.26620","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2606.26620/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2606.26620","created_at":"2026-06-26T01:15:36.437562+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.26620v1","created_at":"2026-06-26T01:15:36.437562+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.26620","created_at":"2026-06-26T01:15:36.437562+00:00"},{"alias_kind":"pith_short_12","alias_value":"BEFU2LW7MUBO","created_at":"2026-06-26T01:15:36.437562+00:00"},{"alias_kind":"pith_short_16","alias_value":"BEFU2LW7MUBON53Y","created_at":"2026-06-26T01:15:36.437562+00:00"},{"alias_kind":"pith_short_8","alias_value":"BEFU2LW7","created_at":"2026-06-26T01:15:36.437562+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX","json":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX.json","graph_json":"https://pith.science/api/pith-number/BEFU2LW7MUBON53YBE2PPLF7EX/graph.json","events_json":"https://pith.science/api/pith-number/BEFU2LW7MUBON53YBE2PPLF7EX/events.json","paper":"https://pith.science/paper/BEFU2LW7"},"agent_actions":{"view_html":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX","download_json":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX.json","view_paper":"https://pith.science/paper/BEFU2LW7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.26620&json=true","fetch_graph":"https://pith.science/api/pith-number/BEFU2LW7MUBON53YBE2PPLF7EX/graph.json","fetch_events":"https://pith.science/api/pith-number/BEFU2LW7MUBON53YBE2PPLF7EX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX/action/storage_attestation","attest_author":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX/action/author_attestation","sign_citation":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX/action/citation_signature","submit_replication":"https://pith.science/pith/BEFU2LW7MUBON53YBE2PPLF7EX/action/replication_record"}},"created_at":"2026-06-26T01:15:36.437562+00:00","updated_at":"2026-06-26T01:15:36.437562+00:00"}