{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:GRKZZCHM3GBTV27BLBYOFP2GFW","short_pith_number":"pith:GRKZZCHM","canonical_record":{"source":{"id":"2506.09477","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-11T07:43:33Z","cross_cats_sorted":[],"title_canon_sha256":"6c78758f18bf3aa728d29202a98dd54c5f828170f170827e0bb81678e6882307","abstract_canon_sha256":"6a7599b9b7d34147b9309f65bc68a792411ebb5d08cd8d345f54cf632f998754"},"schema_version":"1.0"},"canonical_sha256":"34559c88ecd9833aebe15870e2bf462dbf122dec835bf9473b66f47221b18908","source":{"kind":"arxiv","id":"2506.09477","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.09477","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"arxiv_version","alias_value":"2506.09477v1","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.09477","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"pith_short_12","alias_value":"GRKZZCHM3GBT","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"pith_short_16","alias_value":"GRKZZCHM3GBTV27B","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"pith_short_8","alias_value":"GRKZZCHM","created_at":"2026-07-05T11:19:38Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:GRKZZCHM3GBTV27BLBYOFP2GFW","target":"record","payload":{"canonical_record":{"source":{"id":"2506.09477","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-11T07:43:33Z","cross_cats_sorted":[],"title_canon_sha256":"6c78758f18bf3aa728d29202a98dd54c5f828170f170827e0bb81678e6882307","abstract_canon_sha256":"6a7599b9b7d34147b9309f65bc68a792411ebb5d08cd8d345f54cf632f998754"},"schema_version":"1.0"},"canonical_sha256":"34559c88ecd9833aebe15870e2bf462dbf122dec835bf9473b66f47221b18908","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:38.098432Z","signature_b64":"GQf85IX4g7/rGkkekm7zcehdlxqUsZQLaLSrkUwmtna9pdd3z0D67wiJw52gyVR1oLOjRjfW7pAu87ibtQD+CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"34559c88ecd9833aebe15870e2bf462dbf122dec835bf9473b66f47221b18908","last_reissued_at":"2026-07-05T11:19:38.098019Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:38.098019Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2506.09477","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:19:38Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"lxdhvlO7y5mCozdmNkIkHHudFQjqQM2mjsnQ2y6YoZWTlNrcr0uUocMf8O6KjBd9NMaBmRcGJZT51FstetuUBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-11T03:40:15.704573Z"},"content_sha256":"b428b169771a9f58fc6592972ac0f24d5188caff45886ebc9366bff86252bd67","schema_version":"1.0","event_id":"sha256:b428b169771a9f58fc6592972ac0f24d5188caff45886ebc9366bff86252bd67"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:GRKZZCHM3GBTV27BLBYOFP2GFW","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"On a few pitfalls in KL divergence gradient estimation for RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"R\\'emi Munos, Yunhao Tang","submitted_at":"2025-06-11T07:43:33Z","abstract_excerpt":"We point out a few pitfalls in implementing gradient estimation for KL divergence in RL training for LLM, as seen in a number of open source projects and papers. The first major pitfall is to differentiate through the KL estimate as loss functions to minimize KL divergence. We show that such implementations are generally incorrect and do not produce the desired KL gradient. Secondly, we show that some implementations do not account for the sequential nature of the estimation problem and produce a partial gradient at best. We demonstrate the impact of such issues with illustrative tabular and L"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.09477","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/2506.09477/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:19:38Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EtX1+AcaCO12AuA/MshhoUVG1mS15iMpKW1Xo0016CQwUe1NiTm0bsV0x4zfOB72S1gjVL4aPLAczhbcvj/rBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-11T03:40:15.705216Z"},"content_sha256":"3adc86416fbc473dc316d3fd59780e6702df17ff79e9579b1c40515d9c161af7","schema_version":"1.0","event_id":"sha256:3adc86416fbc473dc316d3fd59780e6702df17ff79e9579b1c40515d9c161af7"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/GRKZZCHM3GBTV27BLBYOFP2GFW/bundle.json","state_url":"https://pith.science/pith/GRKZZCHM3GBTV27BLBYOFP2GFW/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/GRKZZCHM3GBTV27BLBYOFP2GFW/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":"2026-08-11T03:40:15Z","links":{"resolver":"https://pith.science/pith/GRKZZCHM3GBTV27BLBYOFP2GFW","bundle":"https://pith.science/pith/GRKZZCHM3GBTV27BLBYOFP2GFW/bundle.json","state":"https://pith.science/pith/GRKZZCHM3GBTV27BLBYOFP2GFW/state.json","well_known_bundle":"https://pith.science/.well-known/pith/GRKZZCHM3GBTV27BLBYOFP2GFW/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:GRKZZCHM3GBTV27BLBYOFP2GFW","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":"6a7599b9b7d34147b9309f65bc68a792411ebb5d08cd8d345f54cf632f998754","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-11T07:43:33Z","title_canon_sha256":"6c78758f18bf3aa728d29202a98dd54c5f828170f170827e0bb81678e6882307"},"schema_version":"1.0","source":{"id":"2506.09477","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.09477","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"arxiv_version","alias_value":"2506.09477v1","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.09477","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"pith_short_12","alias_value":"GRKZZCHM3GBT","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"pith_short_16","alias_value":"GRKZZCHM3GBTV27B","created_at":"2026-07-05T11:19:38Z"},{"alias_kind":"pith_short_8","alias_value":"GRKZZCHM","created_at":"2026-07-05T11:19:38Z"}],"graph_snapshots":[{"event_id":"sha256:3adc86416fbc473dc316d3fd59780e6702df17ff79e9579b1c40515d9c161af7","target":"graph","created_at":"2026-07-05T11:19:38Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2506.09477/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"We point out a few pitfalls in implementing gradient estimation for KL divergence in RL training for LLM, as seen in a number of open source projects and papers. The first major pitfall is to differentiate through the KL estimate as loss functions to minimize KL divergence. We show that such implementations are generally incorrect and do not produce the desired KL gradient. Secondly, we show that some implementations do not account for the sequential nature of the estimation problem and produce a partial gradient at best. We demonstrate the impact of such issues with illustrative tabular and L","authors_text":"R\\'emi Munos, Yunhao Tang","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-11T07:43:33Z","title":"On a few pitfalls in KL divergence gradient estimation for RL"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.09477","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:b428b169771a9f58fc6592972ac0f24d5188caff45886ebc9366bff86252bd67","target":"record","created_at":"2026-07-05T11:19:38Z","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":"6a7599b9b7d34147b9309f65bc68a792411ebb5d08cd8d345f54cf632f998754","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-11T07:43:33Z","title_canon_sha256":"6c78758f18bf3aa728d29202a98dd54c5f828170f170827e0bb81678e6882307"},"schema_version":"1.0","source":{"id":"2506.09477","kind":"arxiv","version":1}},"canonical_sha256":"34559c88ecd9833aebe15870e2bf462dbf122dec835bf9473b66f47221b18908","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"34559c88ecd9833aebe15870e2bf462dbf122dec835bf9473b66f47221b18908","first_computed_at":"2026-07-05T11:19:38.098019Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:19:38.098019Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"GQf85IX4g7/rGkkekm7zcehdlxqUsZQLaLSrkUwmtna9pdd3z0D67wiJw52gyVR1oLOjRjfW7pAu87ibtQD+CQ==","signature_status":"signed_v1","signed_at":"2026-07-05T11:19:38.098432Z","signed_message":"canonical_sha256_bytes"},"source_id":"2506.09477","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b428b169771a9f58fc6592972ac0f24d5188caff45886ebc9366bff86252bd67","sha256:3adc86416fbc473dc316d3fd59780e6702df17ff79e9579b1c40515d9c161af7"],"state_sha256":"fb7eee063f09de2c5572c42cecf2dc212bf21b57f14c85dbbac844fb2c4176f5"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"lbjy7FUDxWw1X3FVmAurLTKlMqLS1F7zbEouaPCmB8zy2ZIK74DDLkbVZUwWfTKMPuZun5K3r09gD8QgEh5fDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-11T03:40:15.713620Z","bundle_sha256":"c86041f0ff5dc2f49b495a1be76fbf448565c18fee9f84ad3ccedfb710d6ec7b"}}