{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:RRAUOTWLJRZBY7TMC2OWH3IE4X","short_pith_number":"pith:RRAUOTWL","canonical_record":{"source":{"id":"2507.14958","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-20T13:36:19Z","cross_cats_sorted":[],"title_canon_sha256":"ac2c3fa5776076fa0ef9e122d373cb5402824b116e833a3601b980072656cef9","abstract_canon_sha256":"3b87d6c9095d0e8e395afa14965640748ca81a73f49d09d4baee7505c34663cf"},"schema_version":"1.0"},"canonical_sha256":"8c41474ecb4c721c7e6c169d63ed04e5ce8b48620c4051a50fe46e0978753dd5","source":{"kind":"arxiv","id":"2507.14958","version":5},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2507.14958","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"arxiv_version","alias_value":"2507.14958v5","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.14958","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"pith_short_12","alias_value":"RRAUOTWLJRZB","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"pith_short_16","alias_value":"RRAUOTWLJRZBY7TM","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"pith_short_8","alias_value":"RRAUOTWL","created_at":"2026-05-26T01:03:13Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:RRAUOTWLJRZBY7TMC2OWH3IE4X","target":"record","payload":{"canonical_record":{"source":{"id":"2507.14958","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-20T13:36:19Z","cross_cats_sorted":[],"title_canon_sha256":"ac2c3fa5776076fa0ef9e122d373cb5402824b116e833a3601b980072656cef9","abstract_canon_sha256":"3b87d6c9095d0e8e395afa14965640748ca81a73f49d09d4baee7505c34663cf"},"schema_version":"1.0"},"canonical_sha256":"8c41474ecb4c721c7e6c169d63ed04e5ce8b48620c4051a50fe46e0978753dd5","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:13.479556Z","signature_b64":"WqNrSckuNzFaQXd7yHHMH4PdWyMoD3tu9U+KSTl5A0cO2TETrqpw0Hhk9mrlRw0t69YM6GzgCHGvzaFNMpQvBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8c41474ecb4c721c7e6c169d63ed04e5ce8b48620c4051a50fe46e0978753dd5","last_reissued_at":"2026-05-26T01:03:13.478924Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:13.478924Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2507.14958","source_version":5,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-26T01:03:13Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"mKFFNki1ScM05BfrNAYVz5PrfjW3Tj8FOa2htikNW1pCQHRhwVSePpZZfv+mafx8ch+welHhj+cfFdHCNaRGBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-06T00:16:30.802977Z"},"content_sha256":"418067f0eab727f034db4bec6f145f3595be00dd73da32a6533ea3a50e66ceb4","schema_version":"1.0","event_id":"sha256:418067f0eab727f034db4bec6f145f3595be00dd73da32a6533ea3a50e66ceb4"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:RRAUOTWLJRZBY7TMC2OWH3IE4X","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"MUR: Momentum Uncertainty guided Reasoning for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Momentum Uncertainty-guided Reasoning lets LLMs allocate thinking budgets adaptively by tracking stepwise uncertainty over time.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Fangzhi Xu, Haiteng Zhao, Hang Yan, Haoran Luo, Jian Zhang, Jun Liu, Luu Anh Tuan, Qika Lin, Rongman Xu, Xiaobao Wu, Yifei Li","submitted_at":"2025-07-20T13:36:19Z","abstract_excerpt":"Large Language Models have achieved impressive performance on reasoning-intensive tasks, yet optimizing their reasoning efficiency remains an open challenge. While Test-Time Scaling (TTS) improves reasoning quality, it often leads to overthinking, wasting tokens on redundant computations. This work investigates how to efficiently and adaptively guide current model' test-time scaling without additional training. Inspired by the concept of momentum in physics, we propose Momentum Uncertainty-guided Reasoning (MUR), which dynamically allocates thinking budgets to critical reasoning steps by track"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"MUR reduces computation by over 45% on average while improving accuracy from 0.33 to 3.46% across MATH-500, AIME24, AIME25, and GPQA-diamond using Qwen3 models of 1.7B, 4B, and 8B sizes.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The paper assumes that stepwise uncertainty can be meaningfully tracked and aggregated over time without additional training and that this aggregation produces a stable signal that generalizes across model sizes and benchmarks (abstract, paragraph on MUR definition and gamma-control).","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MUR dynamically allocates LLM reasoning budgets by aggregating stepwise uncertainty with momentum, cutting tokens over 45% while raising accuracy 0.33-3.46% on MATH-500, AIME, and GPQA.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Momentum Uncertainty-guided Reasoning lets LLMs allocate thinking budgets adaptively by tracking stepwise uncertainty over time.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ee2e0bb382870db95ca7fbf995d2c3646ccfdea9f681cdfa1fae4e917dc970e5"},"source":{"id":"2507.14958","kind":"arxiv","version":5},"verdict":{"id":"4a31c250-c001-4d19-a94d-5b452ffbf2e6","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T03:30:49.721135Z","strongest_claim":"MUR reduces computation by over 45% on average while improving accuracy from 0.33 to 3.46% across MATH-500, AIME24, AIME25, and GPQA-diamond using Qwen3 models of 1.7B, 4B, and 8B sizes.","one_line_summary":"MUR dynamically allocates LLM reasoning budgets by aggregating stepwise uncertainty with momentum, cutting tokens over 45% while raising accuracy 0.33-3.46% on MATH-500, AIME, and GPQA.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The paper assumes that stepwise uncertainty can be meaningfully tracked and aggregated over time without additional training and that this aggregation produces a stable signal that generalizes across model sizes and benchmarks (abstract, paragraph on MUR definition and gamma-control).","pith_extraction_headline":"Momentum Uncertainty-guided Reasoning lets LLMs allocate thinking budgets adaptively by tracking stepwise uncertainty over time."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2507.14958/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":1,"snapshot_sha256":"d8ce2c2db44c5e1f9ff2fb9f45c06d868707cfe3c4889c024c3399ee8dd1efbf"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"4a31c250-c001-4d19-a94d-5b452ffbf2e6"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-26T01:03:13Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Y/rCOmjjnI3yhG0cSsqDcunPy3m7USJEcRaHv8GMFx2Z/GJW42wiCiAJ8RH7MwKWEyFJkpfU47CCJgZ2frCjBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-06T00:16:30.804153Z"},"content_sha256":"6e94b37ffb723dd3adb556a1613fb62a0b3cd4e331a45d61e74a69b3b0ead7e1","schema_version":"1.0","event_id":"sha256:6e94b37ffb723dd3adb556a1613fb62a0b3cd4e331a45d61e74a69b3b0ead7e1"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/RRAUOTWLJRZBY7TMC2OWH3IE4X/bundle.json","state_url":"https://pith.science/pith/RRAUOTWLJRZBY7TMC2OWH3IE4X/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/RRAUOTWLJRZBY7TMC2OWH3IE4X/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":"2026-08-06T00:16:30Z","links":{"resolver":"https://pith.science/pith/RRAUOTWLJRZBY7TMC2OWH3IE4X","bundle":"https://pith.science/pith/RRAUOTWLJRZBY7TMC2OWH3IE4X/bundle.json","state":"https://pith.science/pith/RRAUOTWLJRZBY7TMC2OWH3IE4X/state.json","well_known_bundle":"https://pith.science/.well-known/pith/RRAUOTWLJRZBY7TMC2OWH3IE4X/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:RRAUOTWLJRZBY7TMC2OWH3IE4X","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":"3b87d6c9095d0e8e395afa14965640748ca81a73f49d09d4baee7505c34663cf","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-20T13:36:19Z","title_canon_sha256":"ac2c3fa5776076fa0ef9e122d373cb5402824b116e833a3601b980072656cef9"},"schema_version":"1.0","source":{"id":"2507.14958","kind":"arxiv","version":5}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2507.14958","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"arxiv_version","alias_value":"2507.14958v5","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.14958","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"pith_short_12","alias_value":"RRAUOTWLJRZB","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"pith_short_16","alias_value":"RRAUOTWLJRZBY7TM","created_at":"2026-05-26T01:03:13Z"},{"alias_kind":"pith_short_8","alias_value":"RRAUOTWL","created_at":"2026-05-26T01:03:13Z"}],"graph_snapshots":[{"event_id":"sha256:6e94b37ffb723dd3adb556a1613fb62a0b3cd4e331a45d61e74a69b3b0ead7e1","target":"graph","created_at":"2026-05-26T01:03:13Z","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":"MUR reduces computation by over 45% on average while improving accuracy from 0.33 to 3.46% across MATH-500, AIME24, AIME25, and GPQA-diamond using Qwen3 models of 1.7B, 4B, and 8B sizes."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The paper assumes that stepwise uncertainty can be meaningfully tracked and aggregated over time without additional training and that this aggregation produces a stable signal that generalizes across model sizes and benchmarks (abstract, paragraph on MUR definition and gamma-control)."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"MUR dynamically allocates LLM reasoning budgets by aggregating stepwise uncertainty with momentum, cutting tokens over 45% while raising accuracy 0.33-3.46% on MATH-500, AIME, and GPQA."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Momentum Uncertainty-guided Reasoning lets LLMs allocate thinking budgets adaptively by tracking stepwise uncertainty over time."}],"snapshot_sha256":"ee2e0bb382870db95ca7fbf995d2c3646ccfdea9f681cdfa1fae4e917dc970e5"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"d8ce2c2db44c5e1f9ff2fb9f45c06d868707cfe3c4889c024c3399ee8dd1efbf"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2507.14958/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large Language Models have achieved impressive performance on reasoning-intensive tasks, yet optimizing their reasoning efficiency remains an open challenge. While Test-Time Scaling (TTS) improves reasoning quality, it often leads to overthinking, wasting tokens on redundant computations. This work investigates how to efficiently and adaptively guide current model' test-time scaling without additional training. Inspired by the concept of momentum in physics, we propose Momentum Uncertainty-guided Reasoning (MUR), which dynamically allocates thinking budgets to critical reasoning steps by track","authors_text":"Fangzhi Xu, Haiteng Zhao, Hang Yan, Haoran Luo, Jian Zhang, Jun Liu, Luu Anh Tuan, Qika Lin, Rongman Xu, Xiaobao Wu, Yifei Li","cross_cats":[],"headline":"Momentum Uncertainty-guided Reasoning lets LLMs allocate thinking budgets adaptively by tracking stepwise uncertainty over time.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-20T13:36:19Z","title":"MUR: Momentum Uncertainty guided Reasoning for Large Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.14958","kind":"arxiv","version":5},"verdict":{"created_at":"2026-05-19T03:30:49.721135Z","id":"4a31c250-c001-4d19-a94d-5b452ffbf2e6","model_set":{"reader":"grok-4.3"},"one_line_summary":"MUR dynamically allocates LLM reasoning budgets by aggregating stepwise uncertainty with momentum, cutting tokens over 45% while raising accuracy 0.33-3.46% on MATH-500, AIME, and GPQA.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Momentum Uncertainty-guided Reasoning lets LLMs allocate thinking budgets adaptively by tracking stepwise uncertainty over time.","strongest_claim":"MUR reduces computation by over 45% on average while improving accuracy from 0.33 to 3.46% across MATH-500, AIME24, AIME25, and GPQA-diamond using Qwen3 models of 1.7B, 4B, and 8B sizes.","weakest_assumption":"The paper assumes that stepwise uncertainty can be meaningfully tracked and aggregated over time without additional training and that this aggregation produces a stable signal that generalizes across model sizes and benchmarks (abstract, paragraph on MUR definition and gamma-control)."}},"verdict_id":"4a31c250-c001-4d19-a94d-5b452ffbf2e6"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:418067f0eab727f034db4bec6f145f3595be00dd73da32a6533ea3a50e66ceb4","target":"record","created_at":"2026-05-26T01:03:13Z","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":"3b87d6c9095d0e8e395afa14965640748ca81a73f49d09d4baee7505c34663cf","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-20T13:36:19Z","title_canon_sha256":"ac2c3fa5776076fa0ef9e122d373cb5402824b116e833a3601b980072656cef9"},"schema_version":"1.0","source":{"id":"2507.14958","kind":"arxiv","version":5}},"canonical_sha256":"8c41474ecb4c721c7e6c169d63ed04e5ce8b48620c4051a50fe46e0978753dd5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"8c41474ecb4c721c7e6c169d63ed04e5ce8b48620c4051a50fe46e0978753dd5","first_computed_at":"2026-05-26T01:03:13.478924Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-26T01:03:13.478924Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"WqNrSckuNzFaQXd7yHHMH4PdWyMoD3tu9U+KSTl5A0cO2TETrqpw0Hhk9mrlRw0t69YM6GzgCHGvzaFNMpQvBw==","signature_status":"signed_v1","signed_at":"2026-05-26T01:03:13.479556Z","signed_message":"canonical_sha256_bytes"},"source_id":"2507.14958","source_kind":"arxiv","source_version":5}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:418067f0eab727f034db4bec6f145f3595be00dd73da32a6533ea3a50e66ceb4","sha256:6e94b37ffb723dd3adb556a1613fb62a0b3cd4e331a45d61e74a69b3b0ead7e1"],"state_sha256":"92105cfc60feae2e559f1235d0332fdc965c3e65116cc518eb9d514e997fef91"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"32aadm9mEMHtqwKeFCr92SJeqdqBPNQtlATRUIe6uNa9kar18f0CKT47ubLMrK64pNx2CbODGiKsHfj4H2lGCQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-06T00:16:30.810444Z","bundle_sha256":"484d6d15211a5f5541a00f821954fa385eac1c870fd5c37d7057a1bef291c367"}}