{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:4ZFHMIZZTGRPXXXRVIR6PU4N7N","short_pith_number":"pith:4ZFHMIZZ","schema_version":"1.0","canonical_sha256":"e64a76233999a2fbdef1aa23e7d38dfb6e9610417f51550b4b7f0842f9beccb5","source":{"kind":"arxiv","id":"2506.22396","version":1},"attestation_state":"computed","paper":{"title":"QuickSilver -- Speeding up LLM Inference through Dynamic Token Halting, KV Skipping, Contextual Token Fusion, and Adaptive Matryoshka Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Aditya Kumar Guru, Aman Chadha, Amitava Das, Danush Khanna, Kripabandhu Ghosh, Meetu Malhotra, Rubhav Bahirwani, Srivarshinee Sridhar, Vinija Jain, Zidan Ahmed","submitted_at":"2025-06-27T17:10:32Z","abstract_excerpt":"Inference accounts for the majority of latency and energy consumption in large language model (LLM) deployments, often exceeding 90% of total cost. While training-time efficiency has seen extensive progress, runtime optimization remains a key bottleneck, particularly under autoregressive decoding. Existing approaches -- such as pruning, quantization, early exits, and speculative decoding -- often require retraining, architectural changes, or disrupt decoding compatibility. We introduce QuickSilver, a modular, token-level framework that enables semantic adaptivity at inference time without alte"},"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":"2506.22396","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-27T17:10:32Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"aa7ad9c07b2949fd3428cbc18e1a80504ded218c748caa6501ca21113bdda434","abstract_canon_sha256":"b081069806b1a34573868e68445ef6714ac5522af7c7269670e7f4749fb43373"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:28:26.789851Z","signature_b64":"uuwejOG5fl4I2laG/Ko4tJkT9pNMVjyeYTfUmpQ1WuAErg3nqfJyOyBT6c5/9FK8aOxO6D6egNy5Schxy3gzAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e64a76233999a2fbdef1aa23e7d38dfb6e9610417f51550b4b7f0842f9beccb5","last_reissued_at":"2026-07-05T11:28:26.789239Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:28:26.789239Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"QuickSilver -- Speeding up LLM Inference through Dynamic Token Halting, KV Skipping, Contextual Token Fusion, and Adaptive Matryoshka Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Aditya Kumar Guru, Aman Chadha, Amitava Das, Danush Khanna, Kripabandhu Ghosh, Meetu Malhotra, Rubhav Bahirwani, Srivarshinee Sridhar, Vinija Jain, Zidan Ahmed","submitted_at":"2025-06-27T17:10:32Z","abstract_excerpt":"Inference accounts for the majority of latency and energy consumption in large language model (LLM) deployments, often exceeding 90% of total cost. While training-time efficiency has seen extensive progress, runtime optimization remains a key bottleneck, particularly under autoregressive decoding. Existing approaches -- such as pruning, quantization, early exits, and speculative decoding -- often require retraining, architectural changes, or disrupt decoding compatibility. We introduce QuickSilver, a modular, token-level framework that enables semantic adaptivity at inference time without alte"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.22396","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/2506.22396/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":"2506.22396","created_at":"2026-07-05T11:28:26.789300+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.22396v1","created_at":"2026-07-05T11:28:26.789300+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.22396","created_at":"2026-07-05T11:28:26.789300+00:00"},{"alias_kind":"pith_short_12","alias_value":"4ZFHMIZZTGRP","created_at":"2026-07-05T11:28:26.789300+00:00"},{"alias_kind":"pith_short_16","alias_value":"4ZFHMIZZTGRPXXXR","created_at":"2026-07-05T11:28:26.789300+00:00"},{"alias_kind":"pith_short_8","alias_value":"4ZFHMIZZ","created_at":"2026-07-05T11:28:26.789300+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.18592","citing_title":"Two-dimensional early exit optimisation of LLM inference","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N","json":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N.json","graph_json":"https://pith.science/api/pith-number/4ZFHMIZZTGRPXXXRVIR6PU4N7N/graph.json","events_json":"https://pith.science/api/pith-number/4ZFHMIZZTGRPXXXRVIR6PU4N7N/events.json","paper":"https://pith.science/paper/4ZFHMIZZ"},"agent_actions":{"view_html":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N","download_json":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N.json","view_paper":"https://pith.science/paper/4ZFHMIZZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.22396&json=true","fetch_graph":"https://pith.science/api/pith-number/4ZFHMIZZTGRPXXXRVIR6PU4N7N/graph.json","fetch_events":"https://pith.science/api/pith-number/4ZFHMIZZTGRPXXXRVIR6PU4N7N/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N/action/storage_attestation","attest_author":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N/action/author_attestation","sign_citation":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N/action/citation_signature","submit_replication":"https://pith.science/pith/4ZFHMIZZTGRPXXXRVIR6PU4N7N/action/replication_record"}},"created_at":"2026-07-05T11:28:26.789300+00:00","updated_at":"2026-07-05T11:28:26.789300+00:00"}