{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:W4JH44UZTOPVXJRZMDXN4ZPWTA","short_pith_number":"pith:W4JH44UZ","schema_version":"1.0","canonical_sha256":"b7127e72999b9f5ba63960eede65f6982f8c4ba4682e4d9ed550ad43f903ea41","source":{"kind":"arxiv","id":"2603.22473","version":2},"attestation_state":"computed","paper":{"title":"Component Ablation for Efficient Hybrid Language Model Architectures: Performance, Resilience, and Compression Implications","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Elies Segu\\'i-Mas, Guillermina Tormo-Carb\\'o, Hector Borobia","submitted_at":"2026-03-23T18:41:49Z","abstract_excerpt":"Hybrid language models combine softmax attention with linear-time sequence mechanisms such as state-space or linear-attention layers, but the functional contribution of each component type remains insufficiently characterized. We study component-level ablation in two sub-1B hybrid language models, Qwen3.5-0.8B and Falcon-H1-0.5B, using likelihood-based evaluation, downstream benchmarks, layer-wise interventions, random controls, and representation-level diagnostics.\n  Across the tested models, removing either attention or the alternative sequence-processing pathway substantially degrades perfo"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2603.22473","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-03-23T18:41:49Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"13be42e611a58ce1ee5b6806665f01f388064dd01b96486aaedd665432b7fc2b","abstract_canon_sha256":"bedb72150db626f214208df03e3f484de1fdc29b21349c496520b09cc73a63e0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:04:42.045336Z","signature_b64":"LgKVGdu+C6JhIdLeU2QvZbGSFgbRABf2MfeJJmgkMVuFV4lo5Hkdrmox9LQAFkOAQCALOFlMDs1dC4VvFYvBCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b7127e72999b9f5ba63960eede65f6982f8c4ba4682e4d9ed550ad43f903ea41","last_reissued_at":"2026-06-09T01:04:42.044637Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:04:42.044637Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Component Ablation for Efficient Hybrid Language Model Architectures: Performance, Resilience, and Compression Implications","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Elies Segu\\'i-Mas, Guillermina Tormo-Carb\\'o, Hector Borobia","submitted_at":"2026-03-23T18:41:49Z","abstract_excerpt":"Hybrid language models combine softmax attention with linear-time sequence mechanisms such as state-space or linear-attention layers, but the functional contribution of each component type remains insufficiently characterized. We study component-level ablation in two sub-1B hybrid language models, Qwen3.5-0.8B and Falcon-H1-0.5B, using likelihood-based evaluation, downstream benchmarks, layer-wise interventions, random controls, and representation-level diagnostics.\n  Across the tested models, removing either attention or the alternative sequence-processing pathway substantially degrades perfo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.22473","kind":"arxiv","version":2},"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/2603.22473/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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2603.22473","created_at":"2026-06-09T01:04:42.044696+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.22473v2","created_at":"2026-06-09T01:04:42.044696+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.22473","created_at":"2026-06-09T01:04:42.044696+00:00"},{"alias_kind":"pith_short_12","alias_value":"W4JH44UZTOPV","created_at":"2026-06-09T01:04:42.044696+00:00"},{"alias_kind":"pith_short_16","alias_value":"W4JH44UZTOPVXJRZ","created_at":"2026-06-09T01:04:42.044696+00:00"},{"alias_kind":"pith_short_8","alias_value":"W4JH44UZ","created_at":"2026-06-09T01:04:42.044696+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.15871","citing_title":"Agentic Discovery of Neural Architectures: AIRA-Compose and AIRA-Design","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2604.22127","citing_title":"Where Should LoRA Go? Component-Type Placement in Hybrid Language Models","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01106","citing_title":"Component-Aware Self-Speculative Decoding in Hybrid Language Models","ref_index":19,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA","json":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA.json","graph_json":"https://pith.science/api/pith-number/W4JH44UZTOPVXJRZMDXN4ZPWTA/graph.json","events_json":"https://pith.science/api/pith-number/W4JH44UZTOPVXJRZMDXN4ZPWTA/events.json","paper":"https://pith.science/paper/W4JH44UZ"},"agent_actions":{"view_html":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA","download_json":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA.json","view_paper":"https://pith.science/paper/W4JH44UZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.22473&json=true","fetch_graph":"https://pith.science/api/pith-number/W4JH44UZTOPVXJRZMDXN4ZPWTA/graph.json","fetch_events":"https://pith.science/api/pith-number/W4JH44UZTOPVXJRZMDXN4ZPWTA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA/action/storage_attestation","attest_author":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA/action/author_attestation","sign_citation":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA/action/citation_signature","submit_replication":"https://pith.science/pith/W4JH44UZTOPVXJRZMDXN4ZPWTA/action/replication_record"}},"created_at":"2026-06-09T01:04:42.044696+00:00","updated_at":"2026-06-09T01:04:42.044696+00:00"}