{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:B4SVZODH5N4EP6ZDELNXHWXALU","short_pith_number":"pith:B4SVZODH","canonical_record":{"source":{"id":"2605.00419","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-01T05:31:18Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"0079557779df90dfd96d3e54e2ff66cc28544e4d4120391390920c06bbeab7b7","abstract_canon_sha256":"e3ac34d91bd038345024eb4eaecc48d87b8fa365d8041e8de278aef572e8b3fa"},"schema_version":"1.0"},"canonical_sha256":"0f255cb867eb7847fb2322db73dae05d05f26c0019bb9e00ba4175e076b6d962","source":{"kind":"arxiv","id":"2605.00419","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.00419","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"arxiv_version","alias_value":"2605.00419v2","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.00419","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"pith_short_12","alias_value":"B4SVZODH5N4E","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"pith_short_16","alias_value":"B4SVZODH5N4EP6ZD","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"pith_short_8","alias_value":"B4SVZODH","created_at":"2026-05-26T02:04:11Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:B4SVZODH5N4EP6ZDELNXHWXALU","target":"record","payload":{"canonical_record":{"source":{"id":"2605.00419","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-01T05:31:18Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"0079557779df90dfd96d3e54e2ff66cc28544e4d4120391390920c06bbeab7b7","abstract_canon_sha256":"e3ac34d91bd038345024eb4eaecc48d87b8fa365d8041e8de278aef572e8b3fa"},"schema_version":"1.0"},"canonical_sha256":"0f255cb867eb7847fb2322db73dae05d05f26c0019bb9e00ba4175e076b6d962","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:11.649123Z","signature_b64":"Cm/Qjesx2aJTUY92HCMf7mlpfiAOud0iQC02p5AtHyhQkl+lgUVjK3BW26AuO7GoQwP0LKlfTiPB7GD01kWQBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0f255cb867eb7847fb2322db73dae05d05f26c0019bb9e00ba4175e076b6d962","last_reissued_at":"2026-05-26T02:04:11.647880Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:11.647880Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.00419","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":"2026-05-26T02:04:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"yA8j01Wb1JZYQaD+RYYh/9oX2aus9H+0EvylGQBn9nK6NKRzmXgLCaLSg/0WYDzfvlcKvIbcw9eplJd+7g3vCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-11T22:15:00.908338Z"},"content_sha256":"3522878efc73efc1b6218df9902d3265a52d51212dcf538f2417b6fc1b05e9f3","schema_version":"1.0","event_id":"sha256:3522878efc73efc1b6218df9902d3265a52d51212dcf538f2417b6fc1b05e9f3"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:B4SVZODH5N4EP6ZDELNXHWXALU","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Rethinking LLM Ensembling from the Perspective of Mixture Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"LLM ensembles can sample their averaged distribution by stochastically selecting one model per token.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chonghan Liu, Jiale Fu, Joey Tianyi Zhou, Peijun Wu, Xu Yang, Yuchu Jiang","submitted_at":"2026-05-01T05:31:18Z","abstract_excerpt":"Model ensembling is a well-established technique for improving the performance of machine learning models. Conventionally, this involves averaging the output distributions of multiple models and selecting the most probable label. This idea has been naturally extended to large language models (LLMs), yielding improved performance but incurring substantial computational cost. This inefficiency stems from directly applying conventional ensemble implementation to LLMs, which require a separate forward pass for each model to explicitly compute the ensemble distribution. In this paper, we propose th"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"ME is mathematically equivalent to sampling from the ensemble distribution, but requires invoking only one model, making it 1.78x-2.68x faster than conventional ensemble.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That stochastic per-token model selection according to mixture weights produces token sequences whose distribution exactly matches the averaged ensemble distribution without bias accumulation or need for additional normalization across steps.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ME reinterprets LLM ensembling as a mixture model by sampling a single model stochastically at each token step, matching the ensemble distribution while invoking only one model per step for substantial speed gains.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LLM ensembles can sample their averaged distribution by stochastically selecting one model per token.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"127c7442d41be990d2ca6857fce8f793a00d471a0908f2943ea2dca8e5486954"},"source":{"id":"2605.00419","kind":"arxiv","version":2},"verdict":{"id":"8580942c-7488-4189-90e3-aab4a5866077","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T20:01:44.961132Z","strongest_claim":"ME is mathematically equivalent to sampling from the ensemble distribution, but requires invoking only one model, making it 1.78x-2.68x faster than conventional ensemble.","one_line_summary":"ME reinterprets LLM ensembling as a mixture model by sampling a single model stochastically at each token step, matching the ensemble distribution while invoking only one model per step for substantial speed gains.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That stochastic per-token model selection according to mixture weights produces token sequences whose distribution exactly matches the averaged ensemble distribution without bias accumulation or need for additional normalization across steps.","pith_extraction_headline":"LLM ensembles can sample their averaged distribution by stochastically selecting one model per token."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.00419/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T19:42:57.442958Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T18:10:44.183035Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"dc6ed3a9afa5d10adc3b8a6dcef780cd3b736de76d8d4665ab6247a4847a2839"},"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":"8580942c-7488-4189-90e3-aab4a5866077"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-26T02:04:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"jkdF1LQN2on5Pul2NKA4LXYUv5JWHxSfN3eLqBx8xmOhdM1yqSnuZXz1nLqcfkNzLNsnU/t8zq5aLfmvAWfbCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-11T22:15:00.909305Z"},"content_sha256":"d3f36b739ede0300bb4fb435a13133658d128e8f30820a0c5182a12bca0a4dd1","schema_version":"1.0","event_id":"sha256:d3f36b739ede0300bb4fb435a13133658d128e8f30820a0c5182a12bca0a4dd1"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/B4SVZODH5N4EP6ZDELNXHWXALU/bundle.json","state_url":"https://pith.science/pith/B4SVZODH5N4EP6ZDELNXHWXALU/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/B4SVZODH5N4EP6ZDELNXHWXALU/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":"2026-06-11T22:15:00Z","links":{"resolver":"https://pith.science/pith/B4SVZODH5N4EP6ZDELNXHWXALU","bundle":"https://pith.science/pith/B4SVZODH5N4EP6ZDELNXHWXALU/bundle.json","state":"https://pith.science/pith/B4SVZODH5N4EP6ZDELNXHWXALU/state.json","well_known_bundle":"https://pith.science/.well-known/pith/B4SVZODH5N4EP6ZDELNXHWXALU/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:B4SVZODH5N4EP6ZDELNXHWXALU","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":"e3ac34d91bd038345024eb4eaecc48d87b8fa365d8041e8de278aef572e8b3fa","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-01T05:31:18Z","title_canon_sha256":"0079557779df90dfd96d3e54e2ff66cc28544e4d4120391390920c06bbeab7b7"},"schema_version":"1.0","source":{"id":"2605.00419","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.00419","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"arxiv_version","alias_value":"2605.00419v2","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.00419","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"pith_short_12","alias_value":"B4SVZODH5N4E","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"pith_short_16","alias_value":"B4SVZODH5N4EP6ZD","created_at":"2026-05-26T02:04:11Z"},{"alias_kind":"pith_short_8","alias_value":"B4SVZODH","created_at":"2026-05-26T02:04:11Z"}],"graph_snapshots":[{"event_id":"sha256:d3f36b739ede0300bb4fb435a13133658d128e8f30820a0c5182a12bca0a4dd1","target":"graph","created_at":"2026-05-26T02:04:11Z","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":"ME is mathematically equivalent to sampling from the ensemble distribution, but requires invoking only one model, making it 1.78x-2.68x faster than conventional ensemble."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That stochastic per-token model selection according to mixture weights produces token sequences whose distribution exactly matches the averaged ensemble distribution without bias accumulation or need for additional normalization across steps."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"ME reinterprets LLM ensembling as a mixture model by sampling a single model stochastically at each token step, matching the ensemble distribution while invoking only one model per step for substantial speed gains."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"LLM ensembles can sample their averaged distribution by stochastically selecting one model per token."}],"snapshot_sha256":"127c7442d41be990d2ca6857fce8f793a00d471a0908f2943ea2dca8e5486954"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T19:42:57.442958Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T18:10:44.183035Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.00419/integrity.json","findings":[],"snapshot_sha256":"dc6ed3a9afa5d10adc3b8a6dcef780cd3b736de76d8d4665ab6247a4847a2839","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Model ensembling is a well-established technique for improving the performance of machine learning models. Conventionally, this involves averaging the output distributions of multiple models and selecting the most probable label. This idea has been naturally extended to large language models (LLMs), yielding improved performance but incurring substantial computational cost. This inefficiency stems from directly applying conventional ensemble implementation to LLMs, which require a separate forward pass for each model to explicitly compute the ensemble distribution. In this paper, we propose th","authors_text":"Chonghan Liu, Jiale Fu, Joey Tianyi Zhou, Peijun Wu, Xu Yang, Yuchu Jiang","cross_cats":["cs.CL"],"headline":"LLM ensembles can sample their averaged distribution by stochastically selecting one model per token.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-01T05:31:18Z","title":"Rethinking LLM Ensembling from the Perspective of Mixture Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.00419","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-09T20:01:44.961132Z","id":"8580942c-7488-4189-90e3-aab4a5866077","model_set":{"reader":"grok-4.3"},"one_line_summary":"ME reinterprets LLM ensembling as a mixture model by sampling a single model stochastically at each token step, matching the ensemble distribution while invoking only one model per step for substantial speed gains.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"LLM ensembles can sample their averaged distribution by stochastically selecting one model per token.","strongest_claim":"ME is mathematically equivalent to sampling from the ensemble distribution, but requires invoking only one model, making it 1.78x-2.68x faster than conventional ensemble.","weakest_assumption":"That stochastic per-token model selection according to mixture weights produces token sequences whose distribution exactly matches the averaged ensemble distribution without bias accumulation or need for additional normalization across steps."}},"verdict_id":"8580942c-7488-4189-90e3-aab4a5866077"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:3522878efc73efc1b6218df9902d3265a52d51212dcf538f2417b6fc1b05e9f3","target":"record","created_at":"2026-05-26T02:04:11Z","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":"e3ac34d91bd038345024eb4eaecc48d87b8fa365d8041e8de278aef572e8b3fa","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-01T05:31:18Z","title_canon_sha256":"0079557779df90dfd96d3e54e2ff66cc28544e4d4120391390920c06bbeab7b7"},"schema_version":"1.0","source":{"id":"2605.00419","kind":"arxiv","version":2}},"canonical_sha256":"0f255cb867eb7847fb2322db73dae05d05f26c0019bb9e00ba4175e076b6d962","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"0f255cb867eb7847fb2322db73dae05d05f26c0019bb9e00ba4175e076b6d962","first_computed_at":"2026-05-26T02:04:11.647880Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-26T02:04:11.647880Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Cm/Qjesx2aJTUY92HCMf7mlpfiAOud0iQC02p5AtHyhQkl+lgUVjK3BW26AuO7GoQwP0LKlfTiPB7GD01kWQBg==","signature_status":"signed_v1","signed_at":"2026-05-26T02:04:11.649123Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.00419","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:3522878efc73efc1b6218df9902d3265a52d51212dcf538f2417b6fc1b05e9f3","sha256:d3f36b739ede0300bb4fb435a13133658d128e8f30820a0c5182a12bca0a4dd1"],"state_sha256":"011cff34c19537053a0d654ab9d1f08787eff06ce95947323444556fcfd870be"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"mM25ZfZWKzLtm4RmwCCPC2rjzu4M1y+VTArRkNCiWBo+IkJZtMAFHnryMaA22xBvOm+7JKEpqbCb0RJpzAhRCw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-11T22:15:00.913956Z","bundle_sha256":"da8c6430e0b374cc4c0ada9f0ec7639ec2e7e0002ad56b8fb6ee69f3cb9f786f"}}