{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:SRP7COU5GB3YODBFGUGIILN7V2","short_pith_number":"pith:SRP7COU5","canonical_record":{"source":{"id":"2605.13907","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T03:36:57Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"364ed5714d6e429a86c3f08ba45b900a8118869451a10a9058dc9019f851f158","abstract_canon_sha256":"64b96b9353f36c4ca59c24e29374fefa83ca3be7632760b715f5993f507986ce"},"schema_version":"1.0"},"canonical_sha256":"945ff13a9d3077870c25350c842dbfaeaae04965786a3c765793336b0568f0e5","source":{"kind":"arxiv","id":"2605.13907","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13907","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13907v1","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13907","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"pith_short_12","alias_value":"SRP7COU5GB3Y","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"SRP7COU5GB3YODBF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"SRP7COU5","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:SRP7COU5GB3YODBFGUGIILN7V2","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13907","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T03:36:57Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"364ed5714d6e429a86c3f08ba45b900a8118869451a10a9058dc9019f851f158","abstract_canon_sha256":"64b96b9353f36c4ca59c24e29374fefa83ca3be7632760b715f5993f507986ce"},"schema_version":"1.0"},"canonical_sha256":"945ff13a9d3077870c25350c842dbfaeaae04965786a3c765793336b0568f0e5","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:18.872422Z","signature_b64":"m7mHvBeuvbV9IxXxq7QfZpuqfFjbYuZ+9kS64gwuXzjV9a8ZT1UpL56HbAb09aSLYyg1Qcxv4aWLGiFOloCPAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"945ff13a9d3077870c25350c842dbfaeaae04965786a3c765793336b0568f0e5","last_reissued_at":"2026-05-17T23:39:18.871843Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:18.871843Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13907","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":"2026-05-17T23:39:18Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"jMJTELslwZwyJN1Z3Rej1ppOHNDPwdndpelZKF8qZzgPnLkRpJ3/0rV42eE2LB+Lr27q84rwMNbY+nZxobpdAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T01:17:06.077825Z"},"content_sha256":"f5a13e45ce8b6782c9582ef8a30a69583f921d19966555538180578b8b144ad9","schema_version":"1.0","event_id":"sha256:f5a13e45ce8b6782c9582ef8a30a69583f921d19966555538180578b8b144ad9"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:SRP7COU5GB3YODBFGUGIILN7V2","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"AIS: Adaptive Importance Sampling for Quantized RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Adaptive Importance Sampling corrects non-stationary bias from low-precision rollouts while keeping their speed gains in LLM RL.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Jiajun Zhou, Lingchao Zheng, Ngai Wong, Wei Shao, Yuwei Fan","submitted_at":"2026-05-13T03:36:57Z","abstract_excerpt":"Reinforcement learning (RL) for large language models (LLMs) is dominated by the cost of rollout generation, which has motivated the use of low-precision rollouts (e.g., FP8) paired with a BF16 trainer to improve throughput and reduce memory pressure. This introduces a rollout-training mismatch that biases the policy gradient and can cause training to collapse outright on reasoning benchmarks. We show that the mismatch is non-stationary and acts as a double-edged sword: early in training it provides a stochastic exploration bonus, exposing the gradient to trajectories the trainer would otherwi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"AIS matches the BF16 baseline on most tasks while retaining the 1.5 to 2.76x rollout speedup of FP8 by combining weight reliability, divergence severity, and variance amplification into a per-batch mixing coefficient that interpolates between uncorrected and importance-weighted gradients.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The three real-time diagnostics can be combined into a single mixing coefficient that reliably preserves early-training exploration benefits while suppressing later destabilizing bias across different models, tasks, and training stages without introducing new instabilities.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"AIS adaptively corrects non-stationary policy gradient bias in quantized LLM RL, matching BF16 performance while retaining 1.5-2.76x FP8 rollout speedup.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Adaptive Importance Sampling corrects non-stationary bias from low-precision rollouts while keeping their speed gains in LLM RL.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2fd68b1d443977d73aaceb481c406a1a25ca12a27a183992f92b4c727281f2b5"},"source":{"id":"2605.13907","kind":"arxiv","version":1},"verdict":{"id":"df2c41d6-3fc1-4b2d-a308-ba890af62447","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T03:09:29.288537Z","strongest_claim":"AIS matches the BF16 baseline on most tasks while retaining the 1.5 to 2.76x rollout speedup of FP8 by combining weight reliability, divergence severity, and variance amplification into a per-batch mixing coefficient that interpolates between uncorrected and importance-weighted gradients.","one_line_summary":"AIS adaptively corrects non-stationary policy gradient bias in quantized LLM RL, matching BF16 performance while retaining 1.5-2.76x FP8 rollout speedup.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The three real-time diagnostics can be combined into a single mixing coefficient that reliably preserves early-training exploration benefits while suppressing later destabilizing bias across different models, tasks, and training stages without introducing new instabilities.","pith_extraction_headline":"Adaptive Importance Sampling corrects non-stationary bias from low-precision rollouts while keeping their speed gains in LLM RL."},"references":{"count":33,"sample":[{"doi":"","year":null,"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","ref_index":1,"cited_arxiv_id":"2110.14168","is_internal_anchor":true},{"doi":"","year":null,"title":"DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models","work_id":"07c85cc5-4086-4abc-823b-6d0f4ff784d0","ref_index":2,"cited_arxiv_id":"2512.02556","is_internal_anchor":true},{"doi":"","year":null,"title":"AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning","work_id":"60017c83-cbbd-4807-b8e4-f8149a6aeaf0","ref_index":3,"cited_arxiv_id":"2505.24298","is_internal_anchor":true},{"doi":"","year":null,"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","ref_index":4,"cited_arxiv_id":"2501.12948","is_internal_anchor":true},{"doi":"","year":null,"title":"OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problems","work_id":"19abed3b-0ff6-409b-aded-a50205319aa3","ref_index":5,"cited_arxiv_id":"2402.14008","is_internal_anchor":true}],"resolved_work":33,"snapshot_sha256":"1097be1a26b192962051ee3eb5e922b9174846614d9d282c3c1ddc711a0ca721","internal_anchors":18},"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":"df2c41d6-3fc1-4b2d-a308-ba890af62447"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:18Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Ru77yBdyTY455hIOyyNVyh8q0CiS9gReqfXeQbv0Rbj2Fy5M0D6ERRF6JxmUu97Dktlgw6dAs3vzFkyFwpu4DA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T01:17:06.078361Z"},"content_sha256":"0472ef532cd490e6e7ca72a388509ab38cd4b5465efcde4b5df461ecefcff0e0","schema_version":"1.0","event_id":"sha256:0472ef532cd490e6e7ca72a388509ab38cd4b5465efcde4b5df461ecefcff0e0"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/SRP7COU5GB3YODBFGUGIILN7V2/bundle.json","state_url":"https://pith.science/pith/SRP7COU5GB3YODBFGUGIILN7V2/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/SRP7COU5GB3YODBFGUGIILN7V2/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":"2026-05-28T01:17:06Z","links":{"resolver":"https://pith.science/pith/SRP7COU5GB3YODBFGUGIILN7V2","bundle":"https://pith.science/pith/SRP7COU5GB3YODBFGUGIILN7V2/bundle.json","state":"https://pith.science/pith/SRP7COU5GB3YODBFGUGIILN7V2/state.json","well_known_bundle":"https://pith.science/.well-known/pith/SRP7COU5GB3YODBFGUGIILN7V2/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:SRP7COU5GB3YODBFGUGIILN7V2","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":"64b96b9353f36c4ca59c24e29374fefa83ca3be7632760b715f5993f507986ce","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T03:36:57Z","title_canon_sha256":"364ed5714d6e429a86c3f08ba45b900a8118869451a10a9058dc9019f851f158"},"schema_version":"1.0","source":{"id":"2605.13907","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13907","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13907v1","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13907","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"pith_short_12","alias_value":"SRP7COU5GB3Y","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"SRP7COU5GB3YODBF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"SRP7COU5","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:0472ef532cd490e6e7ca72a388509ab38cd4b5465efcde4b5df461ecefcff0e0","target":"graph","created_at":"2026-05-17T23:39:18Z","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":"AIS matches the BF16 baseline on most tasks while retaining the 1.5 to 2.76x rollout speedup of FP8 by combining weight reliability, divergence severity, and variance amplification into a per-batch mixing coefficient that interpolates between uncorrected and importance-weighted gradients."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The three real-time diagnostics can be combined into a single mixing coefficient that reliably preserves early-training exploration benefits while suppressing later destabilizing bias across different models, tasks, and training stages without introducing new instabilities."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"AIS adaptively corrects non-stationary policy gradient bias in quantized LLM RL, matching BF16 performance while retaining 1.5-2.76x FP8 rollout speedup."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Adaptive Importance Sampling corrects non-stationary bias from low-precision rollouts while keeping their speed gains in LLM RL."}],"snapshot_sha256":"2fd68b1d443977d73aaceb481c406a1a25ca12a27a183992f92b4c727281f2b5"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Reinforcement learning (RL) for large language models (LLMs) is dominated by the cost of rollout generation, which has motivated the use of low-precision rollouts (e.g., FP8) paired with a BF16 trainer to improve throughput and reduce memory pressure. This introduces a rollout-training mismatch that biases the policy gradient and can cause training to collapse outright on reasoning benchmarks. We show that the mismatch is non-stationary and acts as a double-edged sword: early in training it provides a stochastic exploration bonus, exposing the gradient to trajectories the trainer would otherwi","authors_text":"Jiajun Zhou, Lingchao Zheng, Ngai Wong, Wei Shao, Yuwei Fan","cross_cats":["cs.AI","cs.LG"],"headline":"Adaptive Importance Sampling corrects non-stationary bias from low-precision rollouts while keeping their speed gains in LLM RL.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T03:36:57Z","title":"AIS: Adaptive Importance Sampling for Quantized RL"},"references":{"count":33,"internal_anchors":18,"resolved_work":33,"sample":[{"cited_arxiv_id":"2110.14168","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","year":null},{"cited_arxiv_id":"2512.02556","doi":"","is_internal_anchor":true,"ref_index":2,"title":"DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models","work_id":"07c85cc5-4086-4abc-823b-6d0f4ff784d0","year":null},{"cited_arxiv_id":"2505.24298","doi":"","is_internal_anchor":true,"ref_index":3,"title":"AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning","work_id":"60017c83-cbbd-4807-b8e4-f8149a6aeaf0","year":null},{"cited_arxiv_id":"2501.12948","doi":"","is_internal_anchor":true,"ref_index":4,"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","year":null},{"cited_arxiv_id":"2402.14008","doi":"","is_internal_anchor":true,"ref_index":5,"title":"OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problems","work_id":"19abed3b-0ff6-409b-aded-a50205319aa3","year":null}],"snapshot_sha256":"1097be1a26b192962051ee3eb5e922b9174846614d9d282c3c1ddc711a0ca721"},"source":{"id":"2605.13907","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T03:09:29.288537Z","id":"df2c41d6-3fc1-4b2d-a308-ba890af62447","model_set":{"reader":"grok-4.3"},"one_line_summary":"AIS adaptively corrects non-stationary policy gradient bias in quantized LLM RL, matching BF16 performance while retaining 1.5-2.76x FP8 rollout speedup.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Adaptive Importance Sampling corrects non-stationary bias from low-precision rollouts while keeping their speed gains in LLM RL.","strongest_claim":"AIS matches the BF16 baseline on most tasks while retaining the 1.5 to 2.76x rollout speedup of FP8 by combining weight reliability, divergence severity, and variance amplification into a per-batch mixing coefficient that interpolates between uncorrected and importance-weighted gradients.","weakest_assumption":"The three real-time diagnostics can be combined into a single mixing coefficient that reliably preserves early-training exploration benefits while suppressing later destabilizing bias across different models, tasks, and training stages without introducing new instabilities."}},"verdict_id":"df2c41d6-3fc1-4b2d-a308-ba890af62447"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:f5a13e45ce8b6782c9582ef8a30a69583f921d19966555538180578b8b144ad9","target":"record","created_at":"2026-05-17T23:39:18Z","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":"64b96b9353f36c4ca59c24e29374fefa83ca3be7632760b715f5993f507986ce","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T03:36:57Z","title_canon_sha256":"364ed5714d6e429a86c3f08ba45b900a8118869451a10a9058dc9019f851f158"},"schema_version":"1.0","source":{"id":"2605.13907","kind":"arxiv","version":1}},"canonical_sha256":"945ff13a9d3077870c25350c842dbfaeaae04965786a3c765793336b0568f0e5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"945ff13a9d3077870c25350c842dbfaeaae04965786a3c765793336b0568f0e5","first_computed_at":"2026-05-17T23:39:18.871843Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:18.871843Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"m7mHvBeuvbV9IxXxq7QfZpuqfFjbYuZ+9kS64gwuXzjV9a8ZT1UpL56HbAb09aSLYyg1Qcxv4aWLGiFOloCPAw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:18.872422Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13907","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f5a13e45ce8b6782c9582ef8a30a69583f921d19966555538180578b8b144ad9","sha256:0472ef532cd490e6e7ca72a388509ab38cd4b5465efcde4b5df461ecefcff0e0"],"state_sha256":"f8e7d8c9d7182450e6f08e09fe2d7386c6588f6a2557a9ed70c20ba2e87428ab"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"+5d7tqNQMt2pfxuYmaT7kHxxp3/XKlURHtWbKs5U/qnVHqFnUExWfUTdQ3q7iHAqbpwq/UYNXvZWYNYYbFFdAg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-28T01:17:06.080877Z","bundle_sha256":"b730835a8d07a6cfc9533b33834e0e1ac55c85cbc3660e02ff20136309b6b3db"}}