{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:T7PRIAHHEGNOECVUL26FFSFZ4Z","short_pith_number":"pith:T7PRIAHH","canonical_record":{"source":{"id":"2605.13375","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T11:32:03Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2d1a3380f6da5b8e97e08fba4ecd517b81b89b2a512b1ccbac00732ea21f7347","abstract_canon_sha256":"668b20bdedb2dd38cdba55b0c79ff2b87480be5eb068d63703fd1a8a3cecbe48"},"schema_version":"1.0"},"canonical_sha256":"9fdf1400e7219ae20ab45ebc52c8b9e67fe8a9e37215a3affeadd29e38c0b516","source":{"kind":"arxiv","id":"2605.13375","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13375","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13375v1","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13375","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"pith_short_12","alias_value":"T7PRIAHHEGNO","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"T7PRIAHHEGNOECVU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"T7PRIAHH","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:T7PRIAHHEGNOECVUL26FFSFZ4Z","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13375","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T11:32:03Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2d1a3380f6da5b8e97e08fba4ecd517b81b89b2a512b1ccbac00732ea21f7347","abstract_canon_sha256":"668b20bdedb2dd38cdba55b0c79ff2b87480be5eb068d63703fd1a8a3cecbe48"},"schema_version":"1.0"},"canonical_sha256":"9fdf1400e7219ae20ab45ebc52c8b9e67fe8a9e37215a3affeadd29e38c0b516","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:47.918241Z","signature_b64":"Cf2O5CKbpghp4mIRVY8tLzD73SuGIx1/1qj9KSy9mcJabh4Czkz6Wm/7NLI8P/vmtCD64AKRcNZAsqdwscXUDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9fdf1400e7219ae20ab45ebc52c8b9e67fe8a9e37215a3affeadd29e38c0b516","last_reissued_at":"2026-05-18T02:44:47.917854Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:47.917854Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13375","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:47Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"lC4AlVe22XJt7kb4s6bVtE1FPxBG9TNXTYltIBOQYkf4HybQHiUGCcRo9961gYK4M6+ydh3ftvW5RWAptJhgAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-25T21:14:39.470142Z"},"content_sha256":"252aa906c59db739fbe821d9586872edd204b26d8dd975342404585e5b2a2f78","schema_version":"1.0","event_id":"sha256:252aa906c59db739fbe821d9586872edd204b26d8dd975342404585e5b2a2f78"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:T7PRIAHHEGNOECVUL26FFSFZ4Z","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"GRIP-VLM: Group-Relative Importance Pruning for Efficient Vision-Language Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"GRIP-VLM uses reinforcement learning to optimize discrete visual token pruning in VLMs, avoiding suboptimal local minima from gradient relaxations.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Hao Wen, Liang Mi, Lichen Pang, Mingzhe Huang, Shansong Yang, Ting Cao, Weijun Wang, Xin Ding, Yuanchun Li, Yunxin Liu","submitted_at":"2026-05-13T11:32:03Z","abstract_excerpt":"In Vision-Language Models (VLMs), processing a massive number of visual tokens incurs prohibitive computational overhead. While recent training-aware pruning methods attempt to selectively discard redundant tokens, they largely rely on continuous-gradient relaxations. However, visual token pruning is inherently a discrete, non-convex combinatorial problem; consequently, these continuous approximations frequently trap the optimization in sub-optimal local minima, especially under aggressive compression budgets. To overcome this fundamental bottleneck, we propose GRIP-VLM, a Group-Relative Impor"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"GRIP-VLM consistently outperforms heuristic and supervised-learning baselines, achieving a superior Pareto frontier and delivering up to a 15% inference speedup at equal accuracy.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the Group Relative Policy Optimization agent can reliably discover high-quality discrete pruning masks across varying compression budgets without retraining and without the instability typical of RL on combinatorial spaces.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GRIP-VLM applies group-relative policy optimization via reinforcement learning to prune visual tokens in VLMs, yielding up to 15% inference speedup at matched accuracy over prior methods.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"GRIP-VLM uses reinforcement learning to optimize discrete visual token pruning in VLMs, avoiding suboptimal local minima from gradient relaxations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0f8705bb0e12e53541017e23ee834925ef3787f53d5c3f1cdc9cafa0005510e1"},"source":{"id":"2605.13375","kind":"arxiv","version":1},"verdict":{"id":"7199ddd1-4a63-4022-bb64-fbece73cbafa","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:12:10.099597Z","strongest_claim":"GRIP-VLM consistently outperforms heuristic and supervised-learning baselines, achieving a superior Pareto frontier and delivering up to a 15% inference speedup at equal accuracy.","one_line_summary":"GRIP-VLM applies group-relative policy optimization via reinforcement learning to prune visual tokens in VLMs, yielding up to 15% inference speedup at matched accuracy over prior methods.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the Group Relative Policy Optimization agent can reliably discover high-quality discrete pruning masks across varying compression budgets without retraining and without the instability typical of RL on combinatorial spaces.","pith_extraction_headline":"GRIP-VLM uses reinforcement learning to optimize discrete visual token pruning in VLMs, avoiding suboptimal local minima from gradient relaxations."},"references":{"count":42,"sample":[{"doi":"","year":2023,"title":"Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models","work_id":"4819f738-f69f-49dd-8bed-404f647de63a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models","work_id":"a7e3a737-e007-42bc-be89-c4d34c5ee071","ref_index":2,"cited_arxiv_id":"2304.10592","is_internal_anchor":true},{"doi":"","year":2024,"title":"Improved baselines with visual instruction tuning","work_id":"1191d1e0-5e80-4225-a064-e1c04d70e44d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Visual instruction tuning.Advances in neural information processing systems, 2024","work_id":"93eddd32-4e1d-4fb1-aa72-53eb41f6d1f2","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Llama 2: Open Foundation and Fine-Tuned Chat Models","work_id":"68a5177f-d644-44c1-bd4f-4e5278c22f5d","ref_index":5,"cited_arxiv_id":"2307.09288","is_internal_anchor":true}],"resolved_work":42,"snapshot_sha256":"a3d9548cde7690ae65e1aab4675a2c375d1d4f1d743b6dabd1d24314116a5b43","internal_anchors":12},"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"},"verdict_id":"7199ddd1-4a63-4022-bb64-fbece73cbafa"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:47Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ciqvFu22deg6L2W70jt8CT21NlPzPnVzNuQEojQPBYjE0ZZRFREgXeDK4hy8fgk/IaOQnFQ5X2zeV+mRmN58Dw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-25T21:14:39.471167Z"},"content_sha256":"5e2598a1eca5e1e72cc4cc1c414c941b5af104ada302f91344ae0d35a6cc42cb","schema_version":"1.0","event_id":"sha256:5e2598a1eca5e1e72cc4cc1c414c941b5af104ada302f91344ae0d35a6cc42cb"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/T7PRIAHHEGNOECVUL26FFSFZ4Z/bundle.json","state_url":"https://pith.science/pith/T7PRIAHHEGNOECVUL26FFSFZ4Z/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/T7PRIAHHEGNOECVUL26FFSFZ4Z/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-25T21:14:39Z","links":{"resolver":"https://pith.science/pith/T7PRIAHHEGNOECVUL26FFSFZ4Z","bundle":"https://pith.science/pith/T7PRIAHHEGNOECVUL26FFSFZ4Z/bundle.json","state":"https://pith.science/pith/T7PRIAHHEGNOECVUL26FFSFZ4Z/state.json","well_known_bundle":"https://pith.science/.well-known/pith/T7PRIAHHEGNOECVUL26FFSFZ4Z/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:T7PRIAHHEGNOECVUL26FFSFZ4Z","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"668b20bdedb2dd38cdba55b0c79ff2b87480be5eb068d63703fd1a8a3cecbe48","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T11:32:03Z","title_canon_sha256":"2d1a3380f6da5b8e97e08fba4ecd517b81b89b2a512b1ccbac00732ea21f7347"},"schema_version":"1.0","source":{"id":"2605.13375","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13375","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13375v1","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13375","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"pith_short_12","alias_value":"T7PRIAHHEGNO","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"T7PRIAHHEGNOECVU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"T7PRIAHH","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:5e2598a1eca5e1e72cc4cc1c414c941b5af104ada302f91344ae0d35a6cc42cb","target":"graph","created_at":"2026-05-18T02:44:47Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"GRIP-VLM consistently outperforms heuristic and supervised-learning baselines, achieving a superior Pareto frontier and delivering up to a 15% inference speedup at equal accuracy."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the Group Relative Policy Optimization agent can reliably discover high-quality discrete pruning masks across varying compression budgets without retraining and without the instability typical of RL on combinatorial spaces."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"GRIP-VLM applies group-relative policy optimization via reinforcement learning to prune visual tokens in VLMs, yielding up to 15% inference speedup at matched accuracy over prior methods."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"GRIP-VLM uses reinforcement learning to optimize discrete visual token pruning in VLMs, avoiding suboptimal local minima from gradient relaxations."}],"snapshot_sha256":"0f8705bb0e12e53541017e23ee834925ef3787f53d5c3f1cdc9cafa0005510e1"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"In Vision-Language Models (VLMs), processing a massive number of visual tokens incurs prohibitive computational overhead. While recent training-aware pruning methods attempt to selectively discard redundant tokens, they largely rely on continuous-gradient relaxations. However, visual token pruning is inherently a discrete, non-convex combinatorial problem; consequently, these continuous approximations frequently trap the optimization in sub-optimal local minima, especially under aggressive compression budgets. To overcome this fundamental bottleneck, we propose GRIP-VLM, a Group-Relative Impor","authors_text":"Hao Wen, Liang Mi, Lichen Pang, Mingzhe Huang, Shansong Yang, Ting Cao, Weijun Wang, Xin Ding, Yuanchun Li, Yunxin Liu","cross_cats":["cs.AI"],"headline":"GRIP-VLM uses reinforcement learning to optimize discrete visual token pruning in VLMs, avoiding suboptimal local minima from gradient relaxations.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T11:32:03Z","title":"GRIP-VLM: Group-Relative Importance Pruning for Efficient Vision-Language Models"},"references":{"count":42,"internal_anchors":12,"resolved_work":42,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models","work_id":"4819f738-f69f-49dd-8bed-404f647de63a","year":2023},{"cited_arxiv_id":"2304.10592","doi":"","is_internal_anchor":true,"ref_index":2,"title":"MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models","work_id":"a7e3a737-e007-42bc-be89-c4d34c5ee071","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Improved baselines with visual instruction tuning","work_id":"1191d1e0-5e80-4225-a064-e1c04d70e44d","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Visual instruction tuning.Advances in neural information processing systems, 2024","work_id":"93eddd32-4e1d-4fb1-aa72-53eb41f6d1f2","year":2024},{"cited_arxiv_id":"2307.09288","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Llama 2: Open Foundation and Fine-Tuned Chat Models","work_id":"68a5177f-d644-44c1-bd4f-4e5278c22f5d","year":2023}],"snapshot_sha256":"a3d9548cde7690ae65e1aab4675a2c375d1d4f1d743b6dabd1d24314116a5b43"},"source":{"id":"2605.13375","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T19:12:10.099597Z","id":"7199ddd1-4a63-4022-bb64-fbece73cbafa","model_set":{"reader":"grok-4.3"},"one_line_summary":"GRIP-VLM applies group-relative policy optimization via reinforcement learning to prune visual tokens in VLMs, yielding up to 15% inference speedup at matched accuracy over prior methods.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"GRIP-VLM uses reinforcement learning to optimize discrete visual token pruning in VLMs, avoiding suboptimal local minima from gradient relaxations.","strongest_claim":"GRIP-VLM consistently outperforms heuristic and supervised-learning baselines, achieving a superior Pareto frontier and delivering up to a 15% inference speedup at equal accuracy.","weakest_assumption":"That the Group Relative Policy Optimization agent can reliably discover high-quality discrete pruning masks across varying compression budgets without retraining and without the instability typical of RL on combinatorial spaces."}},"verdict_id":"7199ddd1-4a63-4022-bb64-fbece73cbafa"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:252aa906c59db739fbe821d9586872edd204b26d8dd975342404585e5b2a2f78","target":"record","created_at":"2026-05-18T02:44:47Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"668b20bdedb2dd38cdba55b0c79ff2b87480be5eb068d63703fd1a8a3cecbe48","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T11:32:03Z","title_canon_sha256":"2d1a3380f6da5b8e97e08fba4ecd517b81b89b2a512b1ccbac00732ea21f7347"},"schema_version":"1.0","source":{"id":"2605.13375","kind":"arxiv","version":1}},"canonical_sha256":"9fdf1400e7219ae20ab45ebc52c8b9e67fe8a9e37215a3affeadd29e38c0b516","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"9fdf1400e7219ae20ab45ebc52c8b9e67fe8a9e37215a3affeadd29e38c0b516","first_computed_at":"2026-05-18T02:44:47.917854Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:44:47.917854Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Cf2O5CKbpghp4mIRVY8tLzD73SuGIx1/1qj9KSy9mcJabh4Czkz6Wm/7NLI8P/vmtCD64AKRcNZAsqdwscXUDg==","signature_status":"signed_v1","signed_at":"2026-05-18T02:44:47.918241Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13375","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:252aa906c59db739fbe821d9586872edd204b26d8dd975342404585e5b2a2f78","sha256:5e2598a1eca5e1e72cc4cc1c414c941b5af104ada302f91344ae0d35a6cc42cb"],"state_sha256":"7045ae6e626bcc2f5dbea60f774826ebf26663f0d5cf61b46bcc65a8c2910a36"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"d76u7A6kpPEClPUycI48qQ0wlc+O/XQ+Dp1k6shHsayYPBSHEriqtxF3Nv2DMp6Zkb8RLpYbrxqqBnn+0n7+CA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-25T21:14:39.476260Z","bundle_sha256":"761742e2a896330e18006778d171754f7a0dddd14cb082327f817265c57cfc8a"}}