{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Q4OMKE2OBLSIM63NZSTMCF7VFZ","short_pith_number":"pith:Q4OMKE2O","schema_version":"1.0","canonical_sha256":"871cc5134e0ae4867b6dcca6c117f52e4e0569ea0f54aad858d2d4fc23488b16","source":{"kind":"arxiv","id":"2506.05340","version":2},"attestation_state":"computed","paper":{"title":"Exploring Diffusion Transformer Designs via Grafting","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Agrim Gupta, Azalia Mirhoseini, Daniel Y. Fu, Dongjun Kim, Juan Carlos Niebles, Keshigeyan Chandrasegaran, Lea M. Hadzic, Li Fei-Fei, Manling Li, Michael Poli, Stefano Ermon, Stefano Massaroli","submitted_at":"2025-06-05T17:59:40Z","abstract_excerpt":"Designing model architectures requires decisions such as selecting operators (e.g., attention, convolution) and configurations (e.g., depth, width). However, evaluating the impact of these decisions on model quality requires costly pretraining, limiting architectural investigation. Inspired by how new software is built on existing code, we ask: can new architecture designs be studied using pretrained models? To this end, we present grafting, a simple approach for editing pretrained diffusion transformers (DiTs) to materialize new architectures under small compute budgets. Informed by our analy"},"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":"2506.05340","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-06-05T17:59:40Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"6fa09092189dbb992bd7f5eeaf4540256b145b272f640314726e204a092f883f","abstract_canon_sha256":"9dba73f629c551faaa10d12404e9f10c7753479830fbf6017f26e9c6b8d95e7c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:17:25.029745Z","signature_b64":"VCSR0BfGpNnUnig5+8o21TS4h6GVjLk/+53p7OECHgz1rNVji2++Egz1Q8KxMb3G9OKVkWnqSiDLNkQbfFqkAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"871cc5134e0ae4867b6dcca6c117f52e4e0569ea0f54aad858d2d4fc23488b16","last_reissued_at":"2026-07-05T11:17:25.029223Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:17:25.029223Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Exploring Diffusion Transformer Designs via Grafting","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Agrim Gupta, Azalia Mirhoseini, Daniel Y. Fu, Dongjun Kim, Juan Carlos Niebles, Keshigeyan Chandrasegaran, Lea M. Hadzic, Li Fei-Fei, Manling Li, Michael Poli, Stefano Ermon, Stefano Massaroli","submitted_at":"2025-06-05T17:59:40Z","abstract_excerpt":"Designing model architectures requires decisions such as selecting operators (e.g., attention, convolution) and configurations (e.g., depth, width). However, evaluating the impact of these decisions on model quality requires costly pretraining, limiting architectural investigation. Inspired by how new software is built on existing code, we ask: can new architecture designs be studied using pretrained models? To this end, we present grafting, a simple approach for editing pretrained diffusion transformers (DiTs) to materialize new architectures under small compute budgets. Informed by our analy"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.05340","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/2506.05340/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":"2506.05340","created_at":"2026-07-05T11:17:25.029287+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.05340v2","created_at":"2026-07-05T11:17:25.029287+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.05340","created_at":"2026-07-05T11:17:25.029287+00:00"},{"alias_kind":"pith_short_12","alias_value":"Q4OMKE2OBLSI","created_at":"2026-07-05T11:17:25.029287+00:00"},{"alias_kind":"pith_short_16","alias_value":"Q4OMKE2OBLSIM63N","created_at":"2026-07-05T11:17:25.029287+00:00"},{"alias_kind":"pith_short_8","alias_value":"Q4OMKE2O","created_at":"2026-07-05T11:17:25.029287+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2510.18245","citing_title":"Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2602.05449","citing_title":"DisCa: Accelerating Video Diffusion Transformers with Distillation-Compatible Learnable Feature Caching","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2603.06351","citing_title":"DC-DiT: Adaptive Compute and Elastic Inference for Visual Generation via Dynamic Chunking","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14191","citing_title":"CoReDiT: Spatial Coherence-Guided Token Pruning and Reconstruction for Efficient Diffusion Transformers","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06548","citing_title":"Continuous Latent Diffusion Language Model","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02772","citing_title":"Linearizing Vision Transformer with Test-Time Training","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ","json":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ.json","graph_json":"https://pith.science/api/pith-number/Q4OMKE2OBLSIM63NZSTMCF7VFZ/graph.json","events_json":"https://pith.science/api/pith-number/Q4OMKE2OBLSIM63NZSTMCF7VFZ/events.json","paper":"https://pith.science/paper/Q4OMKE2O"},"agent_actions":{"view_html":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ","download_json":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ.json","view_paper":"https://pith.science/paper/Q4OMKE2O","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.05340&json=true","fetch_graph":"https://pith.science/api/pith-number/Q4OMKE2OBLSIM63NZSTMCF7VFZ/graph.json","fetch_events":"https://pith.science/api/pith-number/Q4OMKE2OBLSIM63NZSTMCF7VFZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ/action/storage_attestation","attest_author":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ/action/author_attestation","sign_citation":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ/action/citation_signature","submit_replication":"https://pith.science/pith/Q4OMKE2OBLSIM63NZSTMCF7VFZ/action/replication_record"}},"created_at":"2026-07-05T11:17:25.029287+00:00","updated_at":"2026-07-05T11:17:25.029287+00:00"}