{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:VBXS3Q2EFOBLHXHWFS6FDOC6UT","short_pith_number":"pith:VBXS3Q2E","schema_version":"1.0","canonical_sha256":"a86f2dc3442b82b3dcf62cbc51b85ea4fe8c4c40c90044995b05a221c57a6ccf","source":{"kind":"arxiv","id":"2203.14680","version":3},"attestation_state":"computed","paper":{"title":"Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Avi Caciularu, Kevin Ro Wang, Mor Geva, Yoav Goldberg","submitted_at":"2022-03-28T12:26:00Z","abstract_excerpt":"Transformer-based language models (LMs) are at the core of modern NLP, but their internal prediction construction process is opaque and largely not understood. In this work, we make a substantial step towards unveiling this underlying prediction process, by reverse-engineering the operation of the feed-forward network (FFN) layers, one of the building blocks of transformer models. We view the token representation as a changing distribution over the vocabulary, and the output from each FFN layer as an additive update to that distribution. Then, we analyze the FFN updates in the vocabulary space"},"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":"2203.14680","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-03-28T12:26:00Z","cross_cats_sorted":[],"title_canon_sha256":"7d4e3cac0b89e1a3680448297bbf719913179c1f3d0260da4217ebc62e9acd39","abstract_canon_sha256":"cf7933a31f1e8fd622b6712ec1af41c2915e8a1a8a43589b6123495660b80b78"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:06:01.222159Z","signature_b64":"N+zhMuant+yz9dY440V5XIBw2sWKS93vTK/YFCLrjNaI72RNS66L95wlOd+3/NzHjXTPAThkHDV97Sbk4usrBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a86f2dc3442b82b3dcf62cbc51b85ea4fe8c4c40c90044995b05a221c57a6ccf","last_reissued_at":"2026-07-05T05:06:01.221653Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:06:01.221653Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Avi Caciularu, Kevin Ro Wang, Mor Geva, Yoav Goldberg","submitted_at":"2022-03-28T12:26:00Z","abstract_excerpt":"Transformer-based language models (LMs) are at the core of modern NLP, but their internal prediction construction process is opaque and largely not understood. In this work, we make a substantial step towards unveiling this underlying prediction process, by reverse-engineering the operation of the feed-forward network (FFN) layers, one of the building blocks of transformer models. We view the token representation as a changing distribution over the vocabulary, and the output from each FFN layer as an additive update to that distribution. Then, we analyze the FFN updates in the vocabulary space"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2203.14680","kind":"arxiv","version":3},"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/2203.14680/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":"2203.14680","created_at":"2026-07-05T05:06:01.221717+00:00"},{"alias_kind":"arxiv_version","alias_value":"2203.14680v3","created_at":"2026-07-05T05:06:01.221717+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2203.14680","created_at":"2026-07-05T05:06:01.221717+00:00"},{"alias_kind":"pith_short_12","alias_value":"VBXS3Q2EFOBL","created_at":"2026-07-05T05:06:01.221717+00:00"},{"alias_kind":"pith_short_16","alias_value":"VBXS3Q2EFOBLHXHW","created_at":"2026-07-05T05:06:01.221717+00:00"},{"alias_kind":"pith_short_8","alias_value":"VBXS3Q2E","created_at":"2026-07-05T05:06:01.221717+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06445","citing_title":"Analysis-by-Proxy: Localization Signals in VLMs Operating as Condition Encoders","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2606.17168","citing_title":"RepSelect: Robust LLM Unlearning via Representation Selectivity","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02090","citing_title":"FocusDiT: Masking Queries in Diffusion Transformers for Fine-grained Image Generation","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28273","citing_title":"Vision-Default, Prior-Override: Causal Mechanisms of Perception-Knowledge Conflict in Vision-Language Models","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25384","citing_title":"GeoMathCode: Understanding Interleaved Math-Code Reasoning for Geometry Problem Solving","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11739","citing_title":"Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2510.01685","citing_title":"How Do Language Models Compose Functions?","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2210.07229","citing_title":"Mass-Editing Memory in a Transformer","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":231,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22778","citing_title":"The Spectral Lifecycle of Transformer Training: Transient Compression Waves, Persistent Spectral Gradients, and the Q/K--V Asymmetry","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2303.08112","citing_title":"Eliciting Latent Predictions from Transformers with the Tuned Lens","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22906","citing_title":"Network Edge Inference for Large Language Models: Principles, Techniques, and Opportunities","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13258","citing_title":"Hessian-Enhanced Token Attribution (HETA): Interpreting Autoregressive LLMs","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2308.10248","citing_title":"Steering Language Models With Activation Engineering","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19884","citing_title":"From Signal Degradation to Computation Collapse: Uncovering the Two Failure Modes of LLM Quantization","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05341","citing_title":"Feature Starvation as Geometric Instability in Sparse Autoencoders","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT","json":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT.json","graph_json":"https://pith.science/api/pith-number/VBXS3Q2EFOBLHXHWFS6FDOC6UT/graph.json","events_json":"https://pith.science/api/pith-number/VBXS3Q2EFOBLHXHWFS6FDOC6UT/events.json","paper":"https://pith.science/paper/VBXS3Q2E"},"agent_actions":{"view_html":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT","download_json":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT.json","view_paper":"https://pith.science/paper/VBXS3Q2E","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2203.14680&json=true","fetch_graph":"https://pith.science/api/pith-number/VBXS3Q2EFOBLHXHWFS6FDOC6UT/graph.json","fetch_events":"https://pith.science/api/pith-number/VBXS3Q2EFOBLHXHWFS6FDOC6UT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT/action/storage_attestation","attest_author":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT/action/author_attestation","sign_citation":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT/action/citation_signature","submit_replication":"https://pith.science/pith/VBXS3Q2EFOBLHXHWFS6FDOC6UT/action/replication_record"}},"created_at":"2026-07-05T05:06:01.221717+00:00","updated_at":"2026-07-05T05:06:01.221717+00:00"}