{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:SPDDBVRNS36U3W2AYSQMMOTDSD","short_pith_number":"pith:SPDDBVRN","canonical_record":{"source":{"id":"2511.00066","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-29T08:07:47Z","cross_cats_sorted":[],"title_canon_sha256":"31f0c3e1f267c875ebb9f805cfd7b65021ac2103745a48b4901fd85355cf81c9","abstract_canon_sha256":"69ae622ee2d31532a890af88e4480e6416911004d68999429d7b8d6f7b2cc7d7"},"schema_version":"1.0"},"canonical_sha256":"93c630d62d96fd4ddb40c4a0c63a6390d1fe33a219909cc0a01d9a8d3f0f1d3d","source":{"kind":"arxiv","id":"2511.00066","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2511.00066","created_at":"2026-05-18T03:09:33Z"},{"alias_kind":"arxiv_version","alias_value":"2511.00066v4","created_at":"2026-05-18T03:09:33Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.00066","created_at":"2026-05-18T03:09:33Z"},{"alias_kind":"pith_short_12","alias_value":"SPDDBVRNS36U","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"SPDDBVRNS36U3W2A","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"SPDDBVRN","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:SPDDBVRNS36U3W2AYSQMMOTDSD","target":"record","payload":{"canonical_record":{"source":{"id":"2511.00066","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-29T08:07:47Z","cross_cats_sorted":[],"title_canon_sha256":"31f0c3e1f267c875ebb9f805cfd7b65021ac2103745a48b4901fd85355cf81c9","abstract_canon_sha256":"69ae622ee2d31532a890af88e4480e6416911004d68999429d7b8d6f7b2cc7d7"},"schema_version":"1.0"},"canonical_sha256":"93c630d62d96fd4ddb40c4a0c63a6390d1fe33a219909cc0a01d9a8d3f0f1d3d","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:09:33.672680Z","signature_b64":"zoBZxa80/PVr6s/gMwALgqRurcjDHBdTpf7nCPmGu3wLEg8U2MVv6tLcDbCy/GB/55vjKO//0bgl2iFkLyItCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"93c630d62d96fd4ddb40c4a0c63a6390d1fe33a219909cc0a01d9a8d3f0f1d3d","last_reissued_at":"2026-05-18T03:09:33.672129Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:09:33.672129Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2511.00066","source_version":4,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:33Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"K0qNOmY7L9n9XYu1QqVt/mglhGXEcxSETvNu6AuCUBb8J3O7MfjYIwQ/9nZea0rbspKttqDTXka3mYiqPWa/Bg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-18T18:53:56.670827Z"},"content_sha256":"6be18bbedd2fb57e62fbe87703a323fc132de4d15cb107e7d19bf6be8ba0dab6","schema_version":"1.0","event_id":"sha256:6be18bbedd2fb57e62fbe87703a323fc132de4d15cb107e7d19bf6be8ba0dab6"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:SPDDBVRNS36U3W2AYSQMMOTDSD","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Sharpness-Guided Group Relative Policy Optimization via Probability Shaping","license":"http://creativecommons.org/licenses/by/4.0/","headline":"GRPO-SG is a sharpness-guided token-weighted variant of GRPO that downweights high-gradient tokens to stabilize optimization and improve generalization in reinforcement learning with verifiable rewards.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Linh Ngo Van, Trung Le, Tue Le","submitted_at":"2025-10-29T08:07:47Z","abstract_excerpt":"Reinforcement learning with verifiable rewards (RLVR) has become a practical route to improve large language model reasoning, and Group Relative Policy Optimization (GRPO) is a widely used optimizer in this setting. However, RLVR training is typically performed with limited control over generalization. We revisit GRPO through a robustness-based generalization view, where the generalization loss is upper bounded by a combination of the empirical loss and a sharpness surrogate measured by the gradient norm. Building on this perspective, we propose Sharpness-Guided GRPO (GRPO-SG), a simple token-"},"claims":{"count":3,"items":[{"kind":"strongest_claim","text":"GRPO-SG, a simple token-weighted variant of GRPO, downweights tokens likely to cause overly large gradients, reducing sharp updates and stabilizing optimization, thereby improving generalization in RLVR.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the generalization loss is upper bounded by a combination of the empirical loss and a sharpness surrogate measured by the gradient norm, and that downweighting high-gradient tokens will reliably reduce this sharpness in the RLVR setting for LLMs.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GRPO-SG is a sharpness-guided token-weighted variant of GRPO that downweights high-gradient tokens to stabilize optimization and improve generalization in reinforcement learning with verifiable rewards.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"}],"snapshot_sha256":"9b90ce15ea5ceca22a10db72cd910009dc7570f1d5ea55c60b134457067c5ef3"},"source":{"id":"2511.00066","kind":"arxiv","version":4},"verdict":{"id":"90ca2c5f-5ba7-4e36-81e5-c4d09186bcfa","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-18T03:07:44.282901Z","strongest_claim":"GRPO-SG, a simple token-weighted variant of GRPO, downweights tokens likely to cause overly large gradients, reducing sharp updates and stabilizing optimization, thereby improving generalization in RLVR.","one_line_summary":"GRPO-SG is a sharpness-guided token-weighted variant of GRPO that downweights high-gradient tokens to stabilize optimization and improve generalization in reinforcement learning with verifiable rewards.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the generalization loss is upper bounded by a combination of the empirical loss and a sharpness surrogate measured by the gradient norm, and that downweighting high-gradient tokens will reliably reduce this sharpness in the RLVR setting for LLMs.","pith_extraction_headline":""},"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"},"verdict_id":"90ca2c5f-5ba7-4e36-81e5-c4d09186bcfa"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:33Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"BeY8Hw5zdqOTVEGzo3/AaDCX/yq/Bnn9H+7jJcyCkQlM4kuNckz3v6CWNFMRajCxJxw0zwu5WxeMGOQKRqsrAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-18T18:53:56.671616Z"},"content_sha256":"a28a935b5589c82f382eb03360eedcf075c646f07f6dc431e09a30e1975bbc8d","schema_version":"1.0","event_id":"sha256:a28a935b5589c82f382eb03360eedcf075c646f07f6dc431e09a30e1975bbc8d"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/SPDDBVRNS36U3W2AYSQMMOTDSD/bundle.json","state_url":"https://pith.science/pith/SPDDBVRNS36U3W2AYSQMMOTDSD/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/SPDDBVRNS36U3W2AYSQMMOTDSD/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":"2026-05-18T18:53:56Z","links":{"resolver":"https://pith.science/pith/SPDDBVRNS36U3W2AYSQMMOTDSD","bundle":"https://pith.science/pith/SPDDBVRNS36U3W2AYSQMMOTDSD/bundle.json","state":"https://pith.science/pith/SPDDBVRNS36U3W2AYSQMMOTDSD/state.json","well_known_bundle":"https://pith.science/.well-known/pith/SPDDBVRNS36U3W2AYSQMMOTDSD/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:SPDDBVRNS36U3W2AYSQMMOTDSD","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":"69ae622ee2d31532a890af88e4480e6416911004d68999429d7b8d6f7b2cc7d7","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-29T08:07:47Z","title_canon_sha256":"31f0c3e1f267c875ebb9f805cfd7b65021ac2103745a48b4901fd85355cf81c9"},"schema_version":"1.0","source":{"id":"2511.00066","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2511.00066","created_at":"2026-05-18T03:09:33Z"},{"alias_kind":"arxiv_version","alias_value":"2511.00066v4","created_at":"2026-05-18T03:09:33Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.00066","created_at":"2026-05-18T03:09:33Z"},{"alias_kind":"pith_short_12","alias_value":"SPDDBVRNS36U","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"SPDDBVRNS36U3W2A","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"SPDDBVRN","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:a28a935b5589c82f382eb03360eedcf075c646f07f6dc431e09a30e1975bbc8d","target":"graph","created_at":"2026-05-18T03:09:33Z","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":3,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"GRPO-SG, a simple token-weighted variant of GRPO, downweights tokens likely to cause overly large gradients, reducing sharp updates and stabilizing optimization, thereby improving generalization in RLVR."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the generalization loss is upper bounded by a combination of the empirical loss and a sharpness surrogate measured by the gradient norm, and that downweighting high-gradient tokens will reliably reduce this sharpness in the RLVR setting for LLMs."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"GRPO-SG is a sharpness-guided token-weighted variant of GRPO that downweights high-gradient tokens to stabilize optimization and improve generalization in reinforcement learning with verifiable rewards."}],"snapshot_sha256":"9b90ce15ea5ceca22a10db72cd910009dc7570f1d5ea55c60b134457067c5ef3"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Reinforcement learning with verifiable rewards (RLVR) has become a practical route to improve large language model reasoning, and Group Relative Policy Optimization (GRPO) is a widely used optimizer in this setting. However, RLVR training is typically performed with limited control over generalization. We revisit GRPO through a robustness-based generalization view, where the generalization loss is upper bounded by a combination of the empirical loss and a sharpness surrogate measured by the gradient norm. Building on this perspective, we propose Sharpness-Guided GRPO (GRPO-SG), a simple token-","authors_text":"Linh Ngo Van, Trung Le, Tue Le","cross_cats":[],"headline":"GRPO-SG is a sharpness-guided token-weighted variant of GRPO that downweights high-gradient tokens to stabilize optimization and improve generalization in reinforcement learning with verifiable rewards.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-29T08:07:47Z","title":"Sharpness-Guided Group Relative Policy Optimization via Probability Shaping"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2511.00066","kind":"arxiv","version":4},"verdict":{"created_at":"2026-05-18T03:07:44.282901Z","id":"90ca2c5f-5ba7-4e36-81e5-c4d09186bcfa","model_set":{"reader":"grok-4.3"},"one_line_summary":"GRPO-SG is a sharpness-guided token-weighted variant of GRPO that downweights high-gradient tokens to stabilize optimization and improve generalization in reinforcement learning with verifiable rewards.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"","strongest_claim":"GRPO-SG, a simple token-weighted variant of GRPO, downweights tokens likely to cause overly large gradients, reducing sharp updates and stabilizing optimization, thereby improving generalization in RLVR.","weakest_assumption":"That the generalization loss is upper bounded by a combination of the empirical loss and a sharpness surrogate measured by the gradient norm, and that downweighting high-gradient tokens will reliably reduce this sharpness in the RLVR setting for LLMs."}},"verdict_id":"90ca2c5f-5ba7-4e36-81e5-c4d09186bcfa"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:6be18bbedd2fb57e62fbe87703a323fc132de4d15cb107e7d19bf6be8ba0dab6","target":"record","created_at":"2026-05-18T03:09:33Z","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":"69ae622ee2d31532a890af88e4480e6416911004d68999429d7b8d6f7b2cc7d7","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-29T08:07:47Z","title_canon_sha256":"31f0c3e1f267c875ebb9f805cfd7b65021ac2103745a48b4901fd85355cf81c9"},"schema_version":"1.0","source":{"id":"2511.00066","kind":"arxiv","version":4}},"canonical_sha256":"93c630d62d96fd4ddb40c4a0c63a6390d1fe33a219909cc0a01d9a8d3f0f1d3d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"93c630d62d96fd4ddb40c4a0c63a6390d1fe33a219909cc0a01d9a8d3f0f1d3d","first_computed_at":"2026-05-18T03:09:33.672129Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:33.672129Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"zoBZxa80/PVr6s/gMwALgqRurcjDHBdTpf7nCPmGu3wLEg8U2MVv6tLcDbCy/GB/55vjKO//0bgl2iFkLyItCQ==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:33.672680Z","signed_message":"canonical_sha256_bytes"},"source_id":"2511.00066","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:6be18bbedd2fb57e62fbe87703a323fc132de4d15cb107e7d19bf6be8ba0dab6","sha256:a28a935b5589c82f382eb03360eedcf075c646f07f6dc431e09a30e1975bbc8d"],"state_sha256":"02e09bc0259a0270c88d32a3370839d6c29f02f9e450c2caa97b7062b2fec648"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"VeMJhLDroO3M4DM0djyym/a8wrEPiN0JQTcQkyp2e0vi2BJA/vQicc/GfcIgisgtBtQcputib8ZdSStDuFnvCQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-18T18:53:56.673898Z","bundle_sha256":"a441029cc1e3f56b3a1f07291fc68bf89d8f6282efe7d0031c4d0c00881a5059"}}