{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VB2IORJG3RNLIAR6M5FSV7TRDA","short_pith_number":"pith:VB2IORJG","schema_version":"1.0","canonical_sha256":"a874874526dc5ab4023e674b2afe71181c4397c92e741c908ce3cd20f1561dfd","source":{"kind":"arxiv","id":"2508.21113","version":2},"attestation_state":"computed","paper":{"title":"R-4B: Incentivizing General-Purpose Auto-Thinking Capability in MLLMs via Bi-Mode Annealing and Reinforce Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Bolin Ni, Han Hu, Houwen Peng, Jie Jiang, Qi Yang, Shiming Xiang","submitted_at":"2025-08-28T17:48:19Z","abstract_excerpt":"Multimodal Large Language Models (MLLMs) equipped with step-by-step thinking capabilities have demonstrated remarkable performance on complex reasoning problems. However, this thinking process is redundant for simple problems solvable without complex reasoning. To address this inefficiency, we propose R-4B, an auto-thinking MLLM, which can adaptively decide when to think based on problem complexity. The central idea of R-4B is to empower the model with both thinking and non-thinking capabilities using bi-mode annealing, and apply Bi-mode Policy Optimization (BPO) to improve the model's accurac"},"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":"2508.21113","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-08-28T17:48:19Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"286fbb23b4c888f81f5c86051e864f486483eedd812ed67a4ea6b0741516d075","abstract_canon_sha256":"b544a6117da0f8087a8a0eec6ffa11c298ef0448639aaa49d93d17656a6433fd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:03:32.948766Z","signature_b64":"mru0ZbAWGdSCkySBG0z6bJLeBkLwCHekEC+egVYkI6y4Gmq0nsGgQgfeyQz9dQt5UmO5oGJg3qsCSb8gjAp2CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a874874526dc5ab4023e674b2afe71181c4397c92e741c908ce3cd20f1561dfd","last_reissued_at":"2026-07-05T12:03:32.948256Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:03:32.948256Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"R-4B: Incentivizing General-Purpose Auto-Thinking Capability in MLLMs via Bi-Mode Annealing and Reinforce Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Bolin Ni, Han Hu, Houwen Peng, Jie Jiang, Qi Yang, Shiming Xiang","submitted_at":"2025-08-28T17:48:19Z","abstract_excerpt":"Multimodal Large Language Models (MLLMs) equipped with step-by-step thinking capabilities have demonstrated remarkable performance on complex reasoning problems. However, this thinking process is redundant for simple problems solvable without complex reasoning. To address this inefficiency, we propose R-4B, an auto-thinking MLLM, which can adaptively decide when to think based on problem complexity. The central idea of R-4B is to empower the model with both thinking and non-thinking capabilities using bi-mode annealing, and apply Bi-mode Policy Optimization (BPO) to improve the model's accurac"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.21113","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/2508.21113/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":"2508.21113","created_at":"2026-07-05T12:03:32.948312+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.21113v2","created_at":"2026-07-05T12:03:32.948312+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.21113","created_at":"2026-07-05T12:03:32.948312+00:00"},{"alias_kind":"pith_short_12","alias_value":"VB2IORJG3RNL","created_at":"2026-07-05T12:03:32.948312+00:00"},{"alias_kind":"pith_short_16","alias_value":"VB2IORJG3RNLIAR6","created_at":"2026-07-05T12:03:32.948312+00:00"},{"alias_kind":"pith_short_8","alias_value":"VB2IORJG","created_at":"2026-07-05T12:03:32.948312+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08572","citing_title":"Switch-Reasoner: Learn When to Think in Multitask Mixtures via Reinforcement Learning","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2606.19919","citing_title":"ADaPT: Token-Level Decoupling for Efficient Large Reasoning Models","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA","json":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA.json","graph_json":"https://pith.science/api/pith-number/VB2IORJG3RNLIAR6M5FSV7TRDA/graph.json","events_json":"https://pith.science/api/pith-number/VB2IORJG3RNLIAR6M5FSV7TRDA/events.json","paper":"https://pith.science/paper/VB2IORJG"},"agent_actions":{"view_html":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA","download_json":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA.json","view_paper":"https://pith.science/paper/VB2IORJG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.21113&json=true","fetch_graph":"https://pith.science/api/pith-number/VB2IORJG3RNLIAR6M5FSV7TRDA/graph.json","fetch_events":"https://pith.science/api/pith-number/VB2IORJG3RNLIAR6M5FSV7TRDA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA/action/storage_attestation","attest_author":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA/action/author_attestation","sign_citation":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA/action/citation_signature","submit_replication":"https://pith.science/pith/VB2IORJG3RNLIAR6M5FSV7TRDA/action/replication_record"}},"created_at":"2026-07-05T12:03:32.948312+00:00","updated_at":"2026-07-05T12:03:32.948312+00:00"}