{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:KWOUI2R3UIKPYA7SV6B3QKLMIV","short_pith_number":"pith:KWOUI2R3","schema_version":"1.0","canonical_sha256":"559d446a3ba214fc03f2af83b8296c454f5b7906022020a577967fc59622934b","source":{"kind":"arxiv","id":"2606.01080","version":1},"attestation_state":"computed","paper":{"title":"ThinkSwitch: Context Distillation with LoRA and Weight Interpolation for Specific-Purpose Reasoning Tasks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Dhruv Saini, Rohan Pandey","submitted_at":"2026-05-31T07:57:45Z","abstract_excerpt":"Large language models often improve on difficult tasks by spending inference-time compute on a reasoning trace before producing the final answer. That extra computation can be useful, but it also raises latency, token cost, and deployment complexity. We introduce \\textbf{ThinkSwitch}, a low-compute procedure for co-training paired instruct and thinking checkpoints. Starting from compatible Qwen3-4B instruct and thinking models, each iteration asks the thinking checkpoint to generate answers, removes the reasoning trace, distills the answer-only pairs into the instruct checkpoint with QLoRA, an"},"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":"2606.01080","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-31T07:57:45Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"f3acb593eecd50720e540ae73f50efc012ce2086c59249f9acf6e1c336e8a8b1","abstract_canon_sha256":"7c279d88ec6897acd9ad3662c944273787265908a7c67bc83a64fd0e63855400"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T01:04:20.710607Z","signature_b64":"HzvHK9J+0hwg8CMYohZLj2C/LOK67esPPQ8WwUe4H+H8XUKvhy9ApvX3j51O0YHEwjQ1YrmXjeCQReTfkRCmDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"559d446a3ba214fc03f2af83b8296c454f5b7906022020a577967fc59622934b","last_reissued_at":"2026-06-02T01:04:20.710224Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T01:04:20.710224Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ThinkSwitch: Context Distillation with LoRA and Weight Interpolation for Specific-Purpose Reasoning Tasks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Dhruv Saini, Rohan Pandey","submitted_at":"2026-05-31T07:57:45Z","abstract_excerpt":"Large language models often improve on difficult tasks by spending inference-time compute on a reasoning trace before producing the final answer. That extra computation can be useful, but it also raises latency, token cost, and deployment complexity. We introduce \\textbf{ThinkSwitch}, a low-compute procedure for co-training paired instruct and thinking checkpoints. Starting from compatible Qwen3-4B instruct and thinking models, each iteration asks the thinking checkpoint to generate answers, removes the reasoning trace, distills the answer-only pairs into the instruct checkpoint with QLoRA, an"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.01080","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/2606.01080/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":"2606.01080","created_at":"2026-06-02T01:04:20.710280+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.01080v1","created_at":"2026-06-02T01:04:20.710280+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.01080","created_at":"2026-06-02T01:04:20.710280+00:00"},{"alias_kind":"pith_short_12","alias_value":"KWOUI2R3UIKP","created_at":"2026-06-02T01:04:20.710280+00:00"},{"alias_kind":"pith_short_16","alias_value":"KWOUI2R3UIKPYA7S","created_at":"2026-06-02T01:04:20.710280+00:00"},{"alias_kind":"pith_short_8","alias_value":"KWOUI2R3","created_at":"2026-06-02T01:04:20.710280+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/KWOUI2R3UIKPYA7SV6B3QKLMIV","json":"https://pith.science/pith/KWOUI2R3UIKPYA7SV6B3QKLMIV.json","graph_json":"https://pith.science/api/pith-number/KWOUI2R3UIKPYA7SV6B3QKLMIV/graph.json","events_json":"https://pith.science/api/pith-number/KWOUI2R3UIKPYA7SV6B3QKLMIV/events.json","paper":"https://pith.science/paper/KWOUI2R3"},"agent_actions":{"view_html":"https://pith.science/pith/KWOUI2R3UIKPYA7SV6B3QKLMIV","download_json":"https://pith.science/pith/KWOUI2R3UIKPYA7SV6B3QKLMIV.json","view_paper":"https://pith.science/paper/KWOUI2R3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.01080&json=true","fetch_graph":"https://pith.science/api/pith-number/KWOUI2R3UIKPYA7SV6B3QKLMIV/graph.json","fetch_events":"https://pith.science/api/pith-number/KWOUI2R3UIKPYA7SV6B3QKLMIV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KWOUI2R3UIKPYA7SV6B3QKLMIV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KWOUI2R3UIKPYA7SV6B3QKLMIV/action/storage_attestation","attest_author":"https://pith.science/pith/KWOUI2R3UIKPYA7SV6B3QKLMIV/action/author_attestation","sign_citation":"https://pith.science/pith/KWOUI2R3UIKPYA7SV6B3QKLMIV/action/citation_signature","submit_replication":"https://pith.science/pith/KWOUI2R3UIKPYA7SV6B3QKLMIV/action/replication_record"}},"created_at":"2026-06-02T01:04:20.710280+00:00","updated_at":"2026-06-02T01:04:20.710280+00:00"}