{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:B6HHW5E3MC4CAKOXJOY4CZC2LA","short_pith_number":"pith:B6HHW5E3","schema_version":"1.0","canonical_sha256":"0f8e7b749b60b82029d74bb1c1645a58207364b4a0ef6ea18b566f4911feb18b","source":{"kind":"arxiv","id":"2504.06225","version":1},"attestation_state":"computed","paper":{"title":"Encoder-Decoder Gemma: Improving the Quality-Efficiency Trade-Off via Adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Armand Joulin, Biao Zhang, Fede Lebron, Fedor Moiseev, Joshua Ainslie, Min Ma, Orhan Firat, Paul Suganthan, Surya Bhupatiraju, Zhe Dong","submitted_at":"2025-04-08T17:13:41Z","abstract_excerpt":"While decoder-only large language models (LLMs) have shown impressive results, encoder-decoder models are still widely adopted in real-world applications for their inference efficiency and richer encoder representation. In this paper, we study a novel problem: adapting pretrained decoder-only LLMs to encoder-decoder, with the goal of leveraging the strengths of both approaches to achieve a more favorable quality-efficiency trade-off. We argue that adaptation not only enables inheriting the capability of decoder-only LLMs but also reduces the demand for computation compared to pretraining from "},"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":"2504.06225","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-04-08T17:13:41Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"69195daec797e08a35b3f2896018a15f8712205a722ac6e7ff9b8dccca9896f4","abstract_canon_sha256":"8b07b7bd6d58914449cee320b43e895a8264fcd87b0f87098d98ae9eeadbf6b2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:46:17.537336Z","signature_b64":"x/RZQJgBcD1gyz5Jal0WIyw7vsUIu4SeNZSbBw1uD+TxOqe2W4PGbU5w/HYqOUGcWlVYyh70w3SqMSTZpcakCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0f8e7b749b60b82029d74bb1c1645a58207364b4a0ef6ea18b566f4911feb18b","last_reissued_at":"2026-07-05T10:46:17.536795Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:46:17.536795Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Encoder-Decoder Gemma: Improving the Quality-Efficiency Trade-Off via Adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Armand Joulin, Biao Zhang, Fede Lebron, Fedor Moiseev, Joshua Ainslie, Min Ma, Orhan Firat, Paul Suganthan, Surya Bhupatiraju, Zhe Dong","submitted_at":"2025-04-08T17:13:41Z","abstract_excerpt":"While decoder-only large language models (LLMs) have shown impressive results, encoder-decoder models are still widely adopted in real-world applications for their inference efficiency and richer encoder representation. In this paper, we study a novel problem: adapting pretrained decoder-only LLMs to encoder-decoder, with the goal of leveraging the strengths of both approaches to achieve a more favorable quality-efficiency trade-off. We argue that adaptation not only enables inheriting the capability of decoder-only LLMs but also reduces the demand for computation compared to pretraining from "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.06225","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/2504.06225/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":"2504.06225","created_at":"2026-07-05T10:46:17.536856+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.06225v1","created_at":"2026-07-05T10:46:17.536856+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.06225","created_at":"2026-07-05T10:46:17.536856+00:00"},{"alias_kind":"pith_short_12","alias_value":"B6HHW5E3MC4C","created_at":"2026-07-05T10:46:17.536856+00:00"},{"alias_kind":"pith_short_16","alias_value":"B6HHW5E3MC4CAKOX","created_at":"2026-07-05T10:46:17.536856+00:00"},{"alias_kind":"pith_short_8","alias_value":"B6HHW5E3","created_at":"2026-07-05T10:46:17.536856+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21882","citing_title":"Streaming T5-based Text-to-Speech Synthesis with Limited Lookahead","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16608","citing_title":"To MRL or not to MRL: Text Embeddings are Robust to Truncation Without Matryoshka Learning, Except In Heavy Truncation Scenarios","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08810","citing_title":"Continuous Language Diffusion as a Decoder-Interface Problem","ref_index":82,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18613","citing_title":"SAME: A Semantically-Aligned Music Autoencoder","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21326","citing_title":"MiMIC: Mitigating Visual Modality Collapse in Universal Multimodal Retrieval While Avoiding Semantic Misalignment","ref_index":107,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04291","citing_title":"Leveraging Pretrained Language Models as Energy Functions for Glauber Dynamics Text Diffusion","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA","json":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA.json","graph_json":"https://pith.science/api/pith-number/B6HHW5E3MC4CAKOXJOY4CZC2LA/graph.json","events_json":"https://pith.science/api/pith-number/B6HHW5E3MC4CAKOXJOY4CZC2LA/events.json","paper":"https://pith.science/paper/B6HHW5E3"},"agent_actions":{"view_html":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA","download_json":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA.json","view_paper":"https://pith.science/paper/B6HHW5E3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.06225&json=true","fetch_graph":"https://pith.science/api/pith-number/B6HHW5E3MC4CAKOXJOY4CZC2LA/graph.json","fetch_events":"https://pith.science/api/pith-number/B6HHW5E3MC4CAKOXJOY4CZC2LA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA/action/storage_attestation","attest_author":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA/action/author_attestation","sign_citation":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA/action/citation_signature","submit_replication":"https://pith.science/pith/B6HHW5E3MC4CAKOXJOY4CZC2LA/action/replication_record"}},"created_at":"2026-07-05T10:46:17.536856+00:00","updated_at":"2026-07-05T10:46:17.536856+00:00"}