{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:U6XO6XTRGY4PEKIR5376D5HPUB","short_pith_number":"pith:U6XO6XTR","canonical_record":{"source":{"id":"2605.04062","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-10T15:49:44Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"e95987ca0acee3161604dbd230a255a0153e5fcfd0ad787e9fbb48cb94a878a3","abstract_canon_sha256":"b9c0c6402c9723957641ccdbb7e58a57f288c13d6e7ed0000b02da92b295f19c"},"schema_version":"1.0"},"canonical_sha256":"a7aeef5e713638f22911eeffe1f4efa07da80b970429eeb586ae0fb065a5bcfc","source":{"kind":"arxiv","id":"2605.04062","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.04062","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"arxiv_version","alias_value":"2605.04062v2","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.04062","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"pith_short_12","alias_value":"U6XO6XTRGY4P","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"pith_short_16","alias_value":"U6XO6XTRGY4PEKIR","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"pith_short_8","alias_value":"U6XO6XTR","created_at":"2026-05-22T01:04:04Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:U6XO6XTRGY4PEKIR5376D5HPUB","target":"record","payload":{"canonical_record":{"source":{"id":"2605.04062","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-10T15:49:44Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"e95987ca0acee3161604dbd230a255a0153e5fcfd0ad787e9fbb48cb94a878a3","abstract_canon_sha256":"b9c0c6402c9723957641ccdbb7e58a57f288c13d6e7ed0000b02da92b295f19c"},"schema_version":"1.0"},"canonical_sha256":"a7aeef5e713638f22911eeffe1f4efa07da80b970429eeb586ae0fb065a5bcfc","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T01:04:04.249231Z","signature_b64":"No0uNX5AZKDd+5vMzGsek5vTZmM9nQBzl/WbLSVNznGL1XwVKxB1J+Zi9ESljxp1MRCGr8T0eAAOer9yycO6AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a7aeef5e713638f22911eeffe1f4efa07da80b970429eeb586ae0fb065a5bcfc","last_reissued_at":"2026-05-22T01:04:04.248488Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T01:04:04.248488Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.04062","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-22T01:04:04Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"goNjksUXsUo8FIS+kj47cOCqxb/KNN/zGlkRzxPow1EWxMKtQTaII7OS+rnLniQwTEKgOR+6mq2wCHpuFT6FAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-01T15:59:39.749099Z"},"content_sha256":"34edf2f4a704094098b7ac4cdba63fb0f09a2df189b3d8b03d5ec0970397a271","schema_version":"1.0","event_id":"sha256:34edf2f4a704094098b7ac4cdba63fb0f09a2df189b3d8b03d5ec0970397a271"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:U6XO6XTRGY4PEKIR5376D5HPUB","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"EdgeRazor: A Lightweight Framework for Large Language Models via Mixed-Precision Quantization-Aware Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"EdgeRazor compresses large language models to 1.88 bits while beating all 3-bit methods and using 4-10 times less training compute.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chen Wu, Le-Tong Huang, Nan Li, Shao-Qun Zhang, Shu-Hao Zhang, Xiang-Sheng Deng, Xin-Yi Zou, Zhi-Hua Zhou","submitted_at":"2026-04-10T15:49:44Z","abstract_excerpt":"Quantization has emerged as a mainstream approach for deploying Large Language Models (LLMs) on resource-constrained devices, yet compressing precision below 4-bit typically causes severe performance degradation or prohibitive retraining costs. In this paper, we propose EdgeRazor, a lightweight framework for LLMs via Mixed-Precision Quantization-Aware Distillation. It contains three modules: Structural Quantization with Mixed Precision for fine-grained control of bit-widths, Layer-Adaptive Feature Distillation that dynamically selects the most informative features for alignment, and Entropy-Aw"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"EdgeRazor with 1.88-bit surpasses all contenders with the 3-bit precision, especially outperforms the leading 2-bit PTQ methods by 11.3 points, within a 4-10× lower training budget than the leading QAT approach.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the entropy-aware KL divergence, whose forward-reverse balance is set solely by the teacher's output distribution, provides stable and generalizable training signals across different model families and datasets without requiring per-model hyperparameter search or additional validation sets.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EdgeRazor delivers 1.58-1.88 bit quantized LLMs that outperform 2-3 bit baselines by up to 11.3 points while using 4-10x less training compute than leading QAT methods.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"EdgeRazor compresses large language models to 1.88 bits while beating all 3-bit methods and using 4-10 times less training compute.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5fde3bd799b39c6f9ab13dc44cb820c88948dd2126bd8c13ddbf1b87bbd94471"},"source":{"id":"2605.04062","kind":"arxiv","version":2},"verdict":{"id":"7075f016-020c-41dd-9c63-917657c8a21d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T18:00:49.049966Z","strongest_claim":"EdgeRazor with 1.88-bit surpasses all contenders with the 3-bit precision, especially outperforms the leading 2-bit PTQ methods by 11.3 points, within a 4-10× lower training budget than the leading QAT approach.","one_line_summary":"EdgeRazor delivers 1.58-1.88 bit quantized LLMs that outperform 2-3 bit baselines by up to 11.3 points while using 4-10x less training compute than leading QAT methods.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the entropy-aware KL divergence, whose forward-reverse balance is set solely by the teacher's output distribution, provides stable and generalizable training signals across different model families and datasets without requiring per-model hyperparameter search or additional validation sets.","pith_extraction_headline":"EdgeRazor compresses large language models to 1.88 bits while beating all 3-bit methods and using 4-10 times less training compute."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.04062/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":2,"snapshot_sha256":"0971e30b3ccf4be3acc929478fd3cc90b2820fd91e152fc6e8fa91a2b2e7486c"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"7075f016-020c-41dd-9c63-917657c8a21d"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-22T01:04:04Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"5wHEQIPL0OvILyGeqlIGU6dTI1fS3HAqJsYV6qh9mJt0W2x3Dah1joVtAMvumdRnt7r5yIpLEK7AC1M7LEgqBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-01T15:59:39.749609Z"},"content_sha256":"bda3cd5acdd189dae1e3b326a8723d371e96932ee347a5e9b0a4cc43b9882a21","schema_version":"1.0","event_id":"sha256:bda3cd5acdd189dae1e3b326a8723d371e96932ee347a5e9b0a4cc43b9882a21"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/U6XO6XTRGY4PEKIR5376D5HPUB/bundle.json","state_url":"https://pith.science/pith/U6XO6XTRGY4PEKIR5376D5HPUB/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/U6XO6XTRGY4PEKIR5376D5HPUB/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-01T15:59:39Z","links":{"resolver":"https://pith.science/pith/U6XO6XTRGY4PEKIR5376D5HPUB","bundle":"https://pith.science/pith/U6XO6XTRGY4PEKIR5376D5HPUB/bundle.json","state":"https://pith.science/pith/U6XO6XTRGY4PEKIR5376D5HPUB/state.json","well_known_bundle":"https://pith.science/.well-known/pith/U6XO6XTRGY4PEKIR5376D5HPUB/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:U6XO6XTRGY4PEKIR5376D5HPUB","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"b9c0c6402c9723957641ccdbb7e58a57f288c13d6e7ed0000b02da92b295f19c","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-10T15:49:44Z","title_canon_sha256":"e95987ca0acee3161604dbd230a255a0153e5fcfd0ad787e9fbb48cb94a878a3"},"schema_version":"1.0","source":{"id":"2605.04062","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.04062","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"arxiv_version","alias_value":"2605.04062v2","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.04062","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"pith_short_12","alias_value":"U6XO6XTRGY4P","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"pith_short_16","alias_value":"U6XO6XTRGY4PEKIR","created_at":"2026-05-22T01:04:04Z"},{"alias_kind":"pith_short_8","alias_value":"U6XO6XTR","created_at":"2026-05-22T01:04:04Z"}],"graph_snapshots":[{"event_id":"sha256:bda3cd5acdd189dae1e3b326a8723d371e96932ee347a5e9b0a4cc43b9882a21","target":"graph","created_at":"2026-05-22T01:04:04Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"EdgeRazor with 1.88-bit surpasses all contenders with the 3-bit precision, especially outperforms the leading 2-bit PTQ methods by 11.3 points, within a 4-10× lower training budget than the leading QAT approach."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the entropy-aware KL divergence, whose forward-reverse balance is set solely by the teacher's output distribution, provides stable and generalizable training signals across different model families and datasets without requiring per-model hyperparameter search or additional validation sets."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"EdgeRazor delivers 1.58-1.88 bit quantized LLMs that outperform 2-3 bit baselines by up to 11.3 points while using 4-10x less training compute than leading QAT methods."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"EdgeRazor compresses large language models to 1.88 bits while beating all 3-bit methods and using 4-10 times less training compute."}],"snapshot_sha256":"5fde3bd799b39c6f9ab13dc44cb820c88948dd2126bd8c13ddbf1b87bbd94471"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"0971e30b3ccf4be3acc929478fd3cc90b2820fd91e152fc6e8fa91a2b2e7486c"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2605.04062/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Quantization has emerged as a mainstream approach for deploying Large Language Models (LLMs) on resource-constrained devices, yet compressing precision below 4-bit typically causes severe performance degradation or prohibitive retraining costs. In this paper, we propose EdgeRazor, a lightweight framework for LLMs via Mixed-Precision Quantization-Aware Distillation. It contains three modules: Structural Quantization with Mixed Precision for fine-grained control of bit-widths, Layer-Adaptive Feature Distillation that dynamically selects the most informative features for alignment, and Entropy-Aw","authors_text":"Chen Wu, Le-Tong Huang, Nan Li, Shao-Qun Zhang, Shu-Hao Zhang, Xiang-Sheng Deng, Xin-Yi Zou, Zhi-Hua Zhou","cross_cats":["cs.AI"],"headline":"EdgeRazor compresses large language models to 1.88 bits while beating all 3-bit methods and using 4-10 times less training compute.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-10T15:49:44Z","title":"EdgeRazor: A Lightweight Framework for Large Language Models via Mixed-Precision Quantization-Aware Distillation"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.04062","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-10T18:00:49.049966Z","id":"7075f016-020c-41dd-9c63-917657c8a21d","model_set":{"reader":"grok-4.3"},"one_line_summary":"EdgeRazor delivers 1.58-1.88 bit quantized LLMs that outperform 2-3 bit baselines by up to 11.3 points while using 4-10x less training compute than leading QAT methods.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"EdgeRazor compresses large language models to 1.88 bits while beating all 3-bit methods and using 4-10 times less training compute.","strongest_claim":"EdgeRazor with 1.88-bit surpasses all contenders with the 3-bit precision, especially outperforms the leading 2-bit PTQ methods by 11.3 points, within a 4-10× lower training budget than the leading QAT approach.","weakest_assumption":"That the entropy-aware KL divergence, whose forward-reverse balance is set solely by the teacher's output distribution, provides stable and generalizable training signals across different model families and datasets without requiring per-model hyperparameter search or additional validation sets."}},"verdict_id":"7075f016-020c-41dd-9c63-917657c8a21d"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:34edf2f4a704094098b7ac4cdba63fb0f09a2df189b3d8b03d5ec0970397a271","target":"record","created_at":"2026-05-22T01:04:04Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"b9c0c6402c9723957641ccdbb7e58a57f288c13d6e7ed0000b02da92b295f19c","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-10T15:49:44Z","title_canon_sha256":"e95987ca0acee3161604dbd230a255a0153e5fcfd0ad787e9fbb48cb94a878a3"},"schema_version":"1.0","source":{"id":"2605.04062","kind":"arxiv","version":2}},"canonical_sha256":"a7aeef5e713638f22911eeffe1f4efa07da80b970429eeb586ae0fb065a5bcfc","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a7aeef5e713638f22911eeffe1f4efa07da80b970429eeb586ae0fb065a5bcfc","first_computed_at":"2026-05-22T01:04:04.248488Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-22T01:04:04.248488Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"No0uNX5AZKDd+5vMzGsek5vTZmM9nQBzl/WbLSVNznGL1XwVKxB1J+Zi9ESljxp1MRCGr8T0eAAOer9yycO6AQ==","signature_status":"signed_v1","signed_at":"2026-05-22T01:04:04.249231Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.04062","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:34edf2f4a704094098b7ac4cdba63fb0f09a2df189b3d8b03d5ec0970397a271","sha256:bda3cd5acdd189dae1e3b326a8723d371e96932ee347a5e9b0a4cc43b9882a21"],"state_sha256":"f02303a1912dfba491121447a7fc47b16069f50bea7fa3e58a8e91d6038956de"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"C4m/hL4Rn4ZihspduEUd9EXDqouKWb35Mr8o5rnBKhxVrAq7uhQ7P3ZDXkXnOZUXcZsMMf3LYKmYS+33d0yqBQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-01T15:59:39.751954Z","bundle_sha256":"261b91087a6cc8afbaaef30781c624164d3cfc30727d8b12a2b6fb5a66e53aff"}}