{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XPFPJJWNFRMUCGVHM7LMAP2DGQ","short_pith_number":"pith:XPFPJJWN","schema_version":"1.0","canonical_sha256":"bbcaf4a6cd2c59411aa767d6c03f43340e1e43c4fefda976d0b3826e981ab3e5","source":{"kind":"arxiv","id":"2502.11517","version":2},"attestation_state":"computed","paper":{"title":"Learning to Keep a Promise: Scaling Language Model Decoding Parallelism with Learned Asynchronous Decoding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.DC","cs.LG"],"primary_cat":"cs.CL","authors_text":"Amir Yazdanbakhsh, Blake M. Elias, Ellie Y. Cheng, Jonathan Ragan-Kelley, Michael Carbin, Nikunj Saunshi, Suvinay Subramanian, Tian Jin, Zack Ankner","submitted_at":"2025-02-17T07:39:16Z","abstract_excerpt":"Decoding with autoregressive large language models (LLMs) traditionally occurs sequentially, generating one token after another. An emerging line of work explored parallel decoding by identifying and simultaneously generating semantically independent chunks of LLM responses. However, these techniques rely on hand-crafted heuristics tied to syntactic structures like lists and paragraphs, making them rigid and imprecise. We present PASTA, a learning-based system that teaches LLMs to identify semantic independence and express parallel decoding opportunities in their own responses. At its core are"},"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":"2502.11517","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-02-17T07:39:16Z","cross_cats_sorted":["cs.DC","cs.LG"],"title_canon_sha256":"f6af32ad929385142303e1648cc8053d56336fcac85b625b836d5802fd314f91","abstract_canon_sha256":"c590119430a33b2057faa40fd0904d952fc5b2f54a109b0b6dffbf09acde8efc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-10T01:09:15.530606Z","signature_b64":"gesp5cucUfbngqev/NUlY4eh8X8Eax6G6Gmw7+/PPEcNUFgOgLh38hHvB22/4MGGtL/plPdY5hcKT0XNvVITAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bbcaf4a6cd2c59411aa767d6c03f43340e1e43c4fefda976d0b3826e981ab3e5","last_reissued_at":"2026-06-10T01:09:15.529498Z","signature_status":"signed_v1","first_computed_at":"2026-06-10T01:09:15.529498Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning to Keep a Promise: Scaling Language Model Decoding Parallelism with Learned Asynchronous Decoding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.DC","cs.LG"],"primary_cat":"cs.CL","authors_text":"Amir Yazdanbakhsh, Blake M. Elias, Ellie Y. Cheng, Jonathan Ragan-Kelley, Michael Carbin, Nikunj Saunshi, Suvinay Subramanian, Tian Jin, Zack Ankner","submitted_at":"2025-02-17T07:39:16Z","abstract_excerpt":"Decoding with autoregressive large language models (LLMs) traditionally occurs sequentially, generating one token after another. An emerging line of work explored parallel decoding by identifying and simultaneously generating semantically independent chunks of LLM responses. However, these techniques rely on hand-crafted heuristics tied to syntactic structures like lists and paragraphs, making them rigid and imprecise. We present PASTA, a learning-based system that teaches LLMs to identify semantic independence and express parallel decoding opportunities in their own responses. At its core are"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.11517","kind":"arxiv","version":2},"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/2502.11517/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":"2502.11517","created_at":"2026-06-10T01:09:15.529665+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.11517v2","created_at":"2026-06-10T01:09:15.529665+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.11517","created_at":"2026-06-10T01:09:15.529665+00:00"},{"alias_kind":"pith_short_12","alias_value":"XPFPJJWNFRMU","created_at":"2026-06-10T01:09:15.529665+00:00"},{"alias_kind":"pith_short_16","alias_value":"XPFPJJWNFRMUCGVH","created_at":"2026-06-10T01:09:15.529665+00:00"},{"alias_kind":"pith_short_8","alias_value":"XPFPJJWN","created_at":"2026-06-10T01:09:15.529665+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.06914","citing_title":"Regulating Branch Parallelism in LLM Serving","ref_index":5,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ","json":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ.json","graph_json":"https://pith.science/api/pith-number/XPFPJJWNFRMUCGVHM7LMAP2DGQ/graph.json","events_json":"https://pith.science/api/pith-number/XPFPJJWNFRMUCGVHM7LMAP2DGQ/events.json","paper":"https://pith.science/paper/XPFPJJWN"},"agent_actions":{"view_html":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ","download_json":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ.json","view_paper":"https://pith.science/paper/XPFPJJWN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.11517&json=true","fetch_graph":"https://pith.science/api/pith-number/XPFPJJWNFRMUCGVHM7LMAP2DGQ/graph.json","fetch_events":"https://pith.science/api/pith-number/XPFPJJWNFRMUCGVHM7LMAP2DGQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ/action/storage_attestation","attest_author":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ/action/author_attestation","sign_citation":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ/action/citation_signature","submit_replication":"https://pith.science/pith/XPFPJJWNFRMUCGVHM7LMAP2DGQ/action/replication_record"}},"created_at":"2026-06-10T01:09:15.529665+00:00","updated_at":"2026-06-10T01:09:15.529665+00:00"}