{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:4CD6ILV7F64FQUIQJD352EL4UI","short_pith_number":"pith:4CD6ILV7","schema_version":"1.0","canonical_sha256":"e087e42ebf2fb858511048f7dd117ca23350fbd46eb018216f27aad0143851ee","source":{"kind":"arxiv","id":"2606.18114","version":1},"attestation_state":"computed","paper":{"title":"Ternary Mamba: Grouped Quantization-Aware Training of W1.58A16 State Space Models","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ramprasath Ganesaraja, Sahil Dilip Panse, Swathika N","submitted_at":"2026-06-16T16:18:21Z","abstract_excerpt":"State Space Models (SSMs) such as Mamba-2 offer linear-time inference but their memory footprint limits edge deployment. Prior ternary SSM work (Slender-Mamba) trains from scratch on 150B tokens; we show a pretrained checkpoint suffices, reducing the marginal token budget by 1,000x. Using grouped quantization-aware training (QAT) with knowledge distillation from a frozen FP16 teacher, we compress Mamba-2 1.3B to 3.61x (2,687 to 744 MB) and achieve 48.1% zero-shot accuracy (7-task average) in just 102M tokens (4 GPU-hours, single H100) -- approaching Bi-Mamba's 48.4% (within +/-0.9pp CI). This "},"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":"2606.18114","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-16T16:18:21Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0a3d961b6912e9e020c82be9c10ec45fd59ce177f1840b74c1ca24eb0fe10cac","abstract_canon_sha256":"30325ff12766a5ce97fd6ce68ec248ec5f29cf0d24e5063416871d47fd3627ee"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:10:48.767708Z","signature_b64":"tCi/DuMb+frpBLp0lFxZeuEHr28LZRlrFWqXUhtlKiVa4PX/ClehnOYlboidmnP63CRcrWSJ8PP2YF2bS2rLBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e087e42ebf2fb858511048f7dd117ca23350fbd46eb018216f27aad0143851ee","last_reissued_at":"2026-06-19T16:10:48.767371Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:10:48.767371Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Ternary Mamba: Grouped Quantization-Aware Training of W1.58A16 State Space Models","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ramprasath Ganesaraja, Sahil Dilip Panse, Swathika N","submitted_at":"2026-06-16T16:18:21Z","abstract_excerpt":"State Space Models (SSMs) such as Mamba-2 offer linear-time inference but their memory footprint limits edge deployment. Prior ternary SSM work (Slender-Mamba) trains from scratch on 150B tokens; we show a pretrained checkpoint suffices, reducing the marginal token budget by 1,000x. Using grouped quantization-aware training (QAT) with knowledge distillation from a frozen FP16 teacher, we compress Mamba-2 1.3B to 3.61x (2,687 to 744 MB) and achieve 48.1% zero-shot accuracy (7-task average) in just 102M tokens (4 GPU-hours, single H100) -- approaching Bi-Mamba's 48.4% (within +/-0.9pp CI). This "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.18114","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/2606.18114/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":"2606.18114","created_at":"2026-06-19T16:10:48.767431+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.18114v1","created_at":"2026-06-19T16:10:48.767431+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.18114","created_at":"2026-06-19T16:10:48.767431+00:00"},{"alias_kind":"pith_short_12","alias_value":"4CD6ILV7F64F","created_at":"2026-06-19T16:10:48.767431+00:00"},{"alias_kind":"pith_short_16","alias_value":"4CD6ILV7F64FQUIQ","created_at":"2026-06-19T16:10:48.767431+00:00"},{"alias_kind":"pith_short_8","alias_value":"4CD6ILV7","created_at":"2026-06-19T16:10:48.767431+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI","json":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI.json","graph_json":"https://pith.science/api/pith-number/4CD6ILV7F64FQUIQJD352EL4UI/graph.json","events_json":"https://pith.science/api/pith-number/4CD6ILV7F64FQUIQJD352EL4UI/events.json","paper":"https://pith.science/paper/4CD6ILV7"},"agent_actions":{"view_html":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI","download_json":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI.json","view_paper":"https://pith.science/paper/4CD6ILV7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.18114&json=true","fetch_graph":"https://pith.science/api/pith-number/4CD6ILV7F64FQUIQJD352EL4UI/graph.json","fetch_events":"https://pith.science/api/pith-number/4CD6ILV7F64FQUIQJD352EL4UI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI/action/storage_attestation","attest_author":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI/action/author_attestation","sign_citation":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI/action/citation_signature","submit_replication":"https://pith.science/pith/4CD6ILV7F64FQUIQJD352EL4UI/action/replication_record"}},"created_at":"2026-06-19T16:10:48.767431+00:00","updated_at":"2026-06-19T16:10:48.767431+00:00"}