{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:HRQX7JBAU3OOLK2OLYXFRVDTW5","short_pith_number":"pith:HRQX7JBA","schema_version":"1.0","canonical_sha256":"3c617fa420a6dce5ab4e5e2e58d473b75f89260931d8611ad2c9a9698570b2c3","source":{"kind":"arxiv","id":"2408.15766","version":3},"attestation_state":"computed","paper":{"title":"Learning Harmonized Representations for Speculative Sampling","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Lefan Zhang, Ruiwen Xu, Xiaodan Wang, Yanhua Huang","submitted_at":"2024-08-28T12:59:12Z","abstract_excerpt":"Speculative sampling is a promising approach to accelerate the decoding stage for Large Language Models (LLMs). Recent advancements that leverage target LLM's contextual information, such as hidden states and KV cache, have shown significant practical improvements. However, these approaches suffer from inconsistent context between training and decoding. We also observe another discrepancy between the training and decoding objectives in existing speculative sampling methods. In this work, we propose a solution named HArmonized Speculative Sampling (HASS) that learns harmonized representations t"},"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":"2408.15766","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2024-08-28T12:59:12Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"e4815dc56df49f507d7774691abf92e4241f40a484e91d6b452c345294fbba27","abstract_canon_sha256":"0dfa70ea229103a213781398b58372a36a09064f2c84b18941b9c52aa237f017"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:20:14.788348Z","signature_b64":"qpYkr4WdhO4QlRAkHpvTPWKS4YSsfdocuRSXnq9TX4L5yGncB47+6Ke4siMueP71RZeN6rbUcY1dxEGLhdH9CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3c617fa420a6dce5ab4e5e2e58d473b75f89260931d8611ad2c9a9698570b2c3","last_reissued_at":"2026-07-05T10:20:14.787624Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:20:14.787624Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning Harmonized Representations for Speculative Sampling","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Lefan Zhang, Ruiwen Xu, Xiaodan Wang, Yanhua Huang","submitted_at":"2024-08-28T12:59:12Z","abstract_excerpt":"Speculative sampling is a promising approach to accelerate the decoding stage for Large Language Models (LLMs). Recent advancements that leverage target LLM's contextual information, such as hidden states and KV cache, have shown significant practical improvements. However, these approaches suffer from inconsistent context between training and decoding. We also observe another discrepancy between the training and decoding objectives in existing speculative sampling methods. In this work, we propose a solution named HArmonized Speculative Sampling (HASS) that learns harmonized representations t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.15766","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/2408.15766/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":"2408.15766","created_at":"2026-07-05T10:20:14.787691+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.15766v3","created_at":"2026-07-05T10:20:14.787691+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.15766","created_at":"2026-07-05T10:20:14.787691+00:00"},{"alias_kind":"pith_short_12","alias_value":"HRQX7JBAU3OO","created_at":"2026-07-05T10:20:14.787691+00:00"},{"alias_kind":"pith_short_16","alias_value":"HRQX7JBAU3OOLK2O","created_at":"2026-07-05T10:20:14.787691+00:00"},{"alias_kind":"pith_short_8","alias_value":"HRQX7JBA","created_at":"2026-07-05T10:20:14.787691+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11552","citing_title":"Teaching Diffusion to Speculate Left-to-Right","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00535","citing_title":"DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation","ref_index":77,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07243","citing_title":"SpecBlock: Block-Iterative Speculative Decoding with Dynamic Tree Drafting","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14978","citing_title":"Performance-Driven Policy Optimization for Speculative Decoding with Adaptive Windowing","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2603.01581","citing_title":"KERV: Kinematic-Rectified Speculative Decoding for Embodied VLA Models","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2603.08899","citing_title":"ConFu: Contemplate the Future for Better Speculative Sampling","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09731","citing_title":"SMART: When is it Actually Worth Expanding a Speculative Tree?","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07243","citing_title":"SpecBlock: Block-Iterative Speculative Decoding with Dynamic Tree Drafting","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5","json":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5.json","graph_json":"https://pith.science/api/pith-number/HRQX7JBAU3OOLK2OLYXFRVDTW5/graph.json","events_json":"https://pith.science/api/pith-number/HRQX7JBAU3OOLK2OLYXFRVDTW5/events.json","paper":"https://pith.science/paper/HRQX7JBA"},"agent_actions":{"view_html":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5","download_json":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5.json","view_paper":"https://pith.science/paper/HRQX7JBA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.15766&json=true","fetch_graph":"https://pith.science/api/pith-number/HRQX7JBAU3OOLK2OLYXFRVDTW5/graph.json","fetch_events":"https://pith.science/api/pith-number/HRQX7JBAU3OOLK2OLYXFRVDTW5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5/action/storage_attestation","attest_author":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5/action/author_attestation","sign_citation":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5/action/citation_signature","submit_replication":"https://pith.science/pith/HRQX7JBAU3OOLK2OLYXFRVDTW5/action/replication_record"}},"created_at":"2026-07-05T10:20:14.787691+00:00","updated_at":"2026-07-05T10:20:14.787691+00:00"}