{"id":"3c9d947de21ceb60b92fd0146ce937e40dce3f6114145548488209a4e5cdeed6","title":"Masato Fujitake introduces EnergyCal for loss spikes in sparse MoE training","argument":"Masato Fujitake’s arXiv abstract introduces EnergyCal, a rule for reducing loss spikes during sparse mixture-of-experts (MoE) language-model training with Muon-family optimizers. It attenuates routed-expert updates according to the normalized squared Frobenius norm of finite-step Newton-Schulz directions, without changing the spectral transform or adding persistent optimizer state. On a 4B-parameter MoE model, the abstract reports lower mean thresholded loss-spike burden and improved final geometric-mean perplexity against a development-selected, lower-learning-rate MuonH recipe, with a modest throughput cost. It also reports reduced spike burden and small endpoint improvements with plain Muon at matched nominal learning rate and weight decay. The results are limited to the evaluated setting; the abstract does not establish how the rule performs across other models or training conditions.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports results on a 4B-parameter MoE model and does not establish performance on other model sizes or training conditions. The abstract reports a modest throughput cost but does not quantify it.","speaker":{"name":"Fujitake, Masato","roleAtSource":"Article author"},"source":{"title":"EnergyCal: Taming Loss Spikes in Muon-Family Optimizers for Sparse MoE Training","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.37062","publishedAt":"2026-10-27","dateBasis":"Date basis not provided."},"evidence":[{"text":"Masato Fujitake introduces EnergyCal as a rule to attenuate routed-expert updates during sparse MoE training.","excerpt":"We introduce EnergyCal, a lightweight rule that attenuates routed-expert updates","url":"https://arxiv.org/abs/2610.37062","relationship":"Source-stated"},{"text":"EnergyCal scales updates according to the normalized squared Frobenius norm of finite-step Newton-Schulz directions, while leaving the spectral transform unchanged and adding no persistent optimizer state.","excerpt":"according to the normalized squared Frobenius norm of their finite-step Newton-Schulz directions. EnergyCal leaves the spectral transform unchanged and adds no persistent optimizer state.","url":"https://arxiv.org/abs/2610.37062","relationship":"Source-stated"},{"text":"On a 4B-parameter MoE model, EnergyCal-MuonH reduced mean thresholded loss-spike burden by 99.51% and final geometric-mean perplexity by 4.06% against a development-selected, lower-learning-rate MuonH recipe, with a modest throughput cost.","excerpt":"reduces mean thresholded loss-spike burden by 99.51% and final geometric-mean perplexity by 4.06%, with a modest throughput cost.","url":"https://arxiv.org/abs/2610.37062","relationship":"Source-stated"},{"text":"With plain Muon at matched nominal learning rate and weight decay, the paper reports an average 95.98% reduction in mean spike burden across two tested exponents, accompanied by small endpoint improvements.","excerpt":"the reduction in mean spike burden averages 95.98% across the two tested exponents, accompanied by small endpoint improvements.","url":"https://arxiv.org/abs/2610.37062","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.37062","checkedAt":"2026-11-06T01:04:07.444Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well does EnergyCal perform on other model sizes and training conditions, beyond the evaluated setting?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Masato Fujitake introduces EnergyCal for loss spikes in sparse MoE training","summary":"Masato Fujitake’s arXiv abstract introduces EnergyCal, a rule for reducing loss spikes during sparse mixture-of-experts (MoE) language-model training with Muon-family optimizers. It attenuates routed-expert updates according to the normalized squared Frobenius norm of finite-step Newton-Schulz directions, without changing the spectral transform or adding persistent optimizer state. On a 4B-parameter MoE model, the abstract reports lower mean thresholded loss-spike burden and improved final geometric-mean perplexity against a development-selected, lower-learning-rate MuonH recipe, with a modest throughput cost. It also reports reduced spike burden and small endpoint improvements with plain Muon at matched nominal learning rate and weight decay. The results are limited to the evaluated setting; the abstract does not establish how the rule performs across other models or training conditions.","question":"How well does EnergyCal perform on other model sizes and training conditions, beyond the evaluated setting?","claims":[{"text":"Masato Fujitake introduces EnergyCal as a rule to attenuate routed-expert updates during sparse MoE training.","excerpt":"We introduce EnergyCal, a lightweight rule that attenuates routed-expert updates","support":"source-stated"},{"text":"EnergyCal scales updates according to the normalized squared Frobenius norm of finite-step Newton-Schulz directions, while leaving the spectral transform unchanged and adding no persistent optimizer state.","excerpt":"according to the normalized squared Frobenius norm of their finite-step Newton-Schulz directions. EnergyCal leaves the spectral transform unchanged and adds no persistent optimizer state.","support":"source-stated"},{"text":"On a 4B-parameter MoE model, EnergyCal-MuonH reduced mean thresholded loss-spike burden by 99.51% and final geometric-mean perplexity by 4.06% against a development-selected, lower-learning-rate MuonH recipe, with a modest throughput cost.","excerpt":"reduces mean thresholded loss-spike burden by 99.51% and final geometric-mean perplexity by 4.06%, with a modest throughput cost.","support":"source-stated"},{"text":"With plain Muon at matched nominal learning rate and weight decay, the paper reports an average 95.98% reduction in mean spike burden across two tested exponents, accompanied by small endpoint improvements.","excerpt":"the reduction in mean spike burden averages 95.98% across the two tested exponents, accompanied by small endpoint improvements.","support":"source-stated"}],"limitations":["The abstract reports results on a 4B-parameter MoE model and does not establish performance on other model sizes or training conditions.","The abstract reports a modest throughput cost but does not quantify it."],"mediaId":"261c78181c99e0249054e59cf68572a29479a2c85aaa8e2ba2f5acbfa3bb21ad","correctionReason":null,"id":"3c9d947de21ceb60b92fd0146ce937e40dce3f6114145548488209a4e5cdeed6","publicationId":"fb9567dc-227b-4c8c-81c7-4791f203d59f","version":1,"sourceUrl":"https://arxiv.org/abs/2610.37062","sourceTitle":"EnergyCal: Taming Loss Spikes in Muon-Family Optimizers for Sparse MoE Training","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-27","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-28T08:40:36.985Z","createdAt":"2026-11-06T00:01:32.455Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"261c78181c99e0249054e59cf68572a29479a2c85aaa8e2ba2f5acbfa3bb21ad","kind":"article","url":"https://arxiv.org/abs/2610.37062","sourceUrl":"https://arxiv.org/abs/2610.37062","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-29T23:01:16.656Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.37062","checkedAt":"2026-11-06T01:04:07.444Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"261c78181c99e0249054e59cf68572a29479a2c85aaa8e2ba2f5acbfa3bb21ad","confidence":0.8200000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"971cdaa84dbe91fb269c5e73dcf004d3d54c0ec0e551a16b5d62d19fe3c8d5bc","decidedAt":"2026-10-29T23:01:17.527Z"},"provenance":{"textHash":"14af7ccd35cd1ba0905c46214b63f5c0cbea6d844185f4c8bdfc04275c619c07","observationId":"3b1b2ff9c24e4411f65761ba49ec4d5bb0d20e4ead2f1f854366ffa6a61600cb","retrievedAt":"2026-10-29T23:01:16.426Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"764dac2e-6533-45ec-8f1f-577530ac1b99"},"reviewer":{"model":"gpt-6-luna","runId":"c0d1eca9-fd6e-49fd-8998-12af04f28b3a"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-06T00:01:32.455Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Fujitake, Masato","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Fujitake, Masato","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Fujitake, Masato","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37062","locator":"head/meta[name=author|citation_author]"},"assertionId":"930af4b67a9ca185f6fb57bb000913f4823b1aaad8f759fddf25148e970873f7:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-29T23:01:16.426Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"930af4b67a9ca185f6fb57bb000913f4823b1aaad8f759fddf25148e970873f7","sourceUrl":"https://arxiv.org/abs/2610.37062","contentSha256":"151a12cd74f81332e4741f86f018f578025d807161f697db09dde95a06c58fbb","textHash":"14af7ccd35cd1ba0905c46214b63f5c0cbea6d844185f4c8bdfc04275c619c07","observedAt":"2026-10-29T23:01:16.426Z"}}],"participationCoverage":{"total":1,"returned":1,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.37062","snapshotId":"930af4b67a9ca185f6fb57bb000913f4823b1aaad8f759fddf25148e970873f7","contentSha256":"151a12cd74f81332e4741f86f018f578025d807161f697db09dde95a06c58fbb","textHash":"14af7ccd35cd1ba0905c46214b63f5c0cbea6d844185f4c8bdfc04275c619c07","observedAt":"2026-10-29T23:01:16.426Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/3c9d947de21ceb60b92fd0146ce937e40dce3f6114145548488209a4e5cdeed6/source"}},"canonicalUrl":"https://escapemeridian.com/readings/masato-fujitake-introduces-energycal-for-loss-spikes-in-sparse-3c9d947d","formats":{"html":"/readings/masato-fujitake-introduces-energycal-for-loss-spikes-in-sparse-3c9d947d","markdown":"/readings/masato-fujitake-introduces-energycal-for-loss-spikes-in-sparse-3c9d947d.md","json":"/readings/masato-fujitake-introduces-energycal-for-loss-spikes-in-sparse-3c9d947d.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-mixture-of-experts","name":"Mixture of experts","explanation":"A model design that sends each input to a few specialized sub-networks instead of the whole model, which saves computation.","referenceUrl":"https://developers.google.com/machine-learning/glossary#mixture-of-experts"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"3c9d947de21ceb60b92fd0146ce937e40dce3f6114145548488209a4e5cdeed6","revision":"fb9567dc-227b-4c8c-81c7-4791f203d59f","revisionKind":"publication-id","title":"Masato Fujitake introduces EnergyCal for loss spikes in sparse MoE training","url":"https://escapemeridian.com/readings/masato-fujitake-introduces-energycal-for-loss-spikes-in-sparse-3c9d947d"},"source":{"type":"source","id":"https://arxiv.org/abs/2610.37062","revision":"14af7ccd35cd1ba0905c46214b63f5c0cbea6d844185f4c8bdfc04275c619c07","revisionKind":"sha256-source-text","title":"EnergyCal: Taming Loss Spikes in Muon-Family Optimizers for Sparse MoE Training","url":"https://arxiv.org/abs/2610.37062"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"3c9d947de21ceb60b92fd0146ce937e40dce3f6114145548488209a4e5cdeed6","title":"Masato Fujitake introduces EnergyCal for loss spikes in sparse MoE training","argument":"Masato Fujitake’s arXiv abstract introduces EnergyCal, a rule for reducing loss spikes during sparse mixture-of-experts (MoE) language-model training with Muon-family optimizers. It attenuates routed-expert updates according to the normalized squared Frobenius norm of finite-step Newton-Schulz directions, without changing the spectral transform or adding persistent optimizer state. On a 4B-parameter MoE model, the abstract reports lower mean thresholded loss-spike burden and improved final geometric-mean perplexity against a development-selected, lower-learning-rate MuonH recipe, with a modest throughput cost. It also reports reduced spike burden and small endpoint improvements with plain Muon at matched nominal learning rate and weight decay. The results are limited to the evaluated setting; the abstract does not establish how the rule performs across other models or training conditions.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports results on a 4B-parameter MoE model and does not establish performance on other model sizes or training conditions. The abstract reports a modest throughput cost but does not quantify it.","speaker":{"name":"Fujitake, Masato","roleAtSource":"Article author"},"source":{"title":"EnergyCal: Taming Loss Spikes in Muon-Family Optimizers for Sparse MoE Training","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.37062","publishedAt":"2026-10-27","dateBasis":"Date basis not provided."},"evidence":[{"text":"Masato Fujitake introduces EnergyCal as a rule to attenuate routed-expert updates during sparse MoE training.","excerpt":"We introduce EnergyCal, a lightweight rule that attenuates routed-expert updates","url":"https://arxiv.org/abs/2610.37062","relationship":"Source-stated"},{"text":"EnergyCal scales updates according to the normalized squared Frobenius norm of finite-step Newton-Schulz directions, while leaving the spectral transform unchanged and adding no persistent optimizer state.","excerpt":"according to the normalized squared Frobenius norm of their finite-step Newton-Schulz directions. EnergyCal leaves the spectral transform unchanged and adds no persistent optimizer state.","url":"https://arxiv.org/abs/2610.37062","relationship":"Source-stated"},{"text":"On a 4B-parameter MoE model, EnergyCal-MuonH reduced mean thresholded loss-spike burden by 99.51% and final geometric-mean perplexity by 4.06% against a development-selected, lower-learning-rate MuonH recipe, with a modest throughput cost.","excerpt":"reduces mean thresholded loss-spike burden by 99.51% and final geometric-mean perplexity by 4.06%, with a modest throughput cost.","url":"https://arxiv.org/abs/2610.37062","relationship":"Source-stated"},{"text":"With plain Muon at matched nominal learning rate and weight decay, the paper reports an average 95.98% reduction in mean spike burden across two tested exponents, accompanied by small endpoint improvements.","excerpt":"the reduction in mean spike burden averages 95.98% across the two tested exponents, accompanied by small endpoint improvements.","url":"https://arxiv.org/abs/2610.37062","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.37062","checkedAt":"2026-11-06T01:04:07.444Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well does EnergyCal perform on other model sizes and training conditions, beyond the evaluated setting?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Masato Fujitake introduces EnergyCal for loss spikes in sparse MoE training","summary":"Masato Fujitake’s arXiv abstract introduces EnergyCal, a rule for reducing loss spikes during sparse mixture-of-experts (MoE) language-model training with Muon-family optimizers. It attenuates routed-expert updates according to the normalized squared Frobenius norm of finite-step Newton-Schulz directions, without changing the spectral transform or adding persistent optimizer state. On a 4B-parameter MoE model, the abstract reports lower mean thresholded loss-spike burden and improved final geometric-mean perplexity against a development-selected, lower-learning-rate MuonH recipe, with a modest throughput cost. It also reports reduced spike burden and small endpoint improvements with plain Muon at matched nominal learning rate and weight decay. The results are limited to the evaluated setting; the abstract does not establish how the rule performs across other models or training conditions.","question":"How well does EnergyCal perform on other model sizes and training conditions, beyond the evaluated setting?","claims":[{"text":"Masato Fujitake introduces EnergyCal as a rule to attenuate routed-expert updates during sparse MoE training.","excerpt":"We introduce EnergyCal, a lightweight rule that attenuates routed-expert updates","support":"source-stated"},{"text":"EnergyCal scales updates according to the normalized squared Frobenius norm of finite-step Newton-Schulz directions, while leaving the spectral transform unchanged and adding no persistent optimizer state.","excerpt":"according to the normalized squared Frobenius norm of their finite-step Newton-Schulz directions. EnergyCal leaves the spectral transform unchanged and adds no persistent optimizer state.","support":"source-stated"},{"text":"On a 4B-parameter MoE model, EnergyCal-MuonH reduced mean thresholded loss-spike burden by 99.51% and final geometric-mean perplexity by 4.06% against a development-selected, lower-learning-rate MuonH recipe, with a modest throughput cost.","excerpt":"reduces mean thresholded loss-spike burden by 99.51% and final geometric-mean perplexity by 4.06%, with a modest throughput cost.","support":"source-stated"},{"text":"With plain Muon at matched nominal learning rate and weight decay, the paper reports an average 95.98% reduction in mean spike burden across two tested exponents, accompanied by small endpoint improvements.","excerpt":"the reduction in mean spike burden averages 95.98% across the two tested exponents, accompanied by small endpoint improvements.","support":"source-stated"}],"limitations":["The abstract reports results on a 4B-parameter MoE model and does not establish performance on other model sizes or training conditions.","The abstract reports a modest throughput cost but does not quantify it."],"mediaId":"261c78181c99e0249054e59cf68572a29479a2c85aaa8e2ba2f5acbfa3bb21ad","correctionReason":null,"id":"3c9d947de21ceb60b92fd0146ce937e40dce3f6114145548488209a4e5cdeed6","publicationId":"fb9567dc-227b-4c8c-81c7-4791f203d59f","version":1,"sourceUrl":"https://arxiv.org/abs/2610.37062","sourceTitle":"EnergyCal: Taming Loss Spikes in Muon-Family Optimizers for Sparse MoE Training","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-27","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-28T08:40:36.985Z","createdAt":"2026-11-06T00:01:32.455Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"261c78181c99e0249054e59cf68572a29479a2c85aaa8e2ba2f5acbfa3bb21ad","kind":"article","url":"https://arxiv.org/abs/2610.37062","sourceUrl":"https://arxiv.org/abs/2610.37062","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-29T23:01:16.656Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.37062","checkedAt":"2026-11-06T01:04:07.444Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"261c78181c99e0249054e59cf68572a29479a2c85aaa8e2ba2f5acbfa3bb21ad","confidence":0.8200000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"971cdaa84dbe91fb269c5e73dcf004d3d54c0ec0e551a16b5d62d19fe3c8d5bc","decidedAt":"2026-10-29T23:01:17.527Z"},"provenance":{"textHash":"14af7ccd35cd1ba0905c46214b63f5c0cbea6d844185f4c8bdfc04275c619c07","observationId":"3b1b2ff9c24e4411f65761ba49ec4d5bb0d20e4ead2f1f854366ffa6a61600cb","retrievedAt":"2026-10-29T23:01:16.426Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"764dac2e-6533-45ec-8f1f-577530ac1b99"},"reviewer":{"model":"gpt-6-luna","runId":"c0d1eca9-fd6e-49fd-8998-12af04f28b3a"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-06T00:01:32.455Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Fujitake, Masato","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Fujitake, Masato","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Fujitake, Masato","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37062","locator":"head/meta[name=author|citation_author]"},"assertionId":"930af4b67a9ca185f6fb57bb000913f4823b1aaad8f759fddf25148e970873f7:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-29T23:01:16.426Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"930af4b67a9ca185f6fb57bb000913f4823b1aaad8f759fddf25148e970873f7","sourceUrl":"https://arxiv.org/abs/2610.37062","contentSha256":"151a12cd74f81332e4741f86f018f578025d807161f697db09dde95a06c58fbb","textHash":"14af7ccd35cd1ba0905c46214b63f5c0cbea6d844185f4c8bdfc04275c619c07","observedAt":"2026-10-29T23:01:16.426Z"}}],"participationCoverage":{"total":1,"returned":1,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.37062","snapshotId":"930af4b67a9ca185f6fb57bb000913f4823b1aaad8f759fddf25148e970873f7","contentSha256":"151a12cd74f81332e4741f86f018f578025d807161f697db09dde95a06c58fbb","textHash":"14af7ccd35cd1ba0905c46214b63f5c0cbea6d844185f4c8bdfc04275c619c07","observedAt":"2026-10-29T23:01:16.426Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/3c9d947de21ceb60b92fd0146ce937e40dce3f6114145548488209a4e5cdeed6/source"}}}],"audits":[{"publicationId":"fb9567dc-227b-4c8c-81c7-4791f203d59f","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft correctly attributes the paper and its claims to Masato Fujitake's arXiv abstract. Results are framed as what the abstract reports, not as established fact or Escape Meridian's view.","excerpt":"Authors:Masato Fujitake"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful but hard for a non-specialist to follow. It never says plainly that the paper is about stabilizing training, which the abstract's conclusion states. It also leaves MuonH unexplained and omits the Hyperball link. Smallest repair: add a sentence that the author argues this stabilizes training both with and without Hyperball (the H in MuonH) in the evaluated setting. Add tooltips for loss spike, Muon, MoE and perplexity. The title and question are fair.","excerpt":"These findings support stabilization both with and without Hyperball in the evaluated setting."},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-10-27 matches the v1 submission date. The later feed date was not used.","excerpt":"[Submitted on 27 Oct 2026]"},{"dimension":"support","status":"pass","explanation":"Every claim is backed by a verbatim abstract excerpt, including the 99.51%, 4.06% and 95.98% figures. Their comparison conditions are preserved: the development-selected lower-learning-rate MuonH recipe, and matched nominal learning rate and weight decay for plain Muon. The limitations follow from the single 4B model and the unquantified throughput cost. No benefits are invented.","excerpt":"We evaluate it on a 4B-parameter MoE model."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-06T00:01:32.455Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}