{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:QFTSVFZBL2VWQMCRSXHZCMJE6Y","short_pith_number":"pith:QFTSVFZB","schema_version":"1.0","canonical_sha256":"81672a97215eab68305195cf913124f63ae20f1024ac0d4dc806faab03ba377a","source":{"kind":"arxiv","id":"2505.22375","version":2},"attestation_state":"computed","paper":{"title":"Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognition","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bin Wang, Chong Zhu, Dacheng Tao, Dong Li, Fei Mi, Hailin Hu, Hanting Chen, Haoyu Wang, Jinpeng Li, Kai Han, Kaikai Song, Lin Li, Mingjian Zhu, Quan He, Wei He, Xinghao Chen, Xueyu Wu, Xu He, Yasheng Wang, Yehui Tang, Yifei Fu, Yu Luo, Yunhe Wang (and Other Contributors), Zhenni Bi","submitted_at":"2025-05-28T14:03:02Z","abstract_excerpt":"This work presents Pangu Embedded, an efficient Large Language Model (LLM) reasoner developed on Ascend Neural Processing Units (NPUs), featuring flexible fast and slow thinking capabilities. Pangu Embedded addresses the significant computational costs and inference latency challenges prevalent in existing reasoning-optimized LLMs. We propose a two-stage training framework for its construction. In Stage 1, the model is finetuned via an iterative distillation process, incorporating inter-iteration model merging to effectively aggregate complementary knowledge. This is followed by reinforcement "},"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":"2505.22375","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-28T14:03:02Z","cross_cats_sorted":[],"title_canon_sha256":"b132456006e49561e08eb7ede26d3069770a39de537d4ca095d8abe5cfbb6e68","abstract_canon_sha256":"af4ab042ed6a77bc7afbcfa61de3a051f0868054a7abd39a9d98a65132207190"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:11:48.108527Z","signature_b64":"gCkByxKQxWxeiVh84ZFrYfLdv0s80MuL/cb/Ca7sF+yWzPXM/53xG8ApfthnbitsjyVzrNmeUY6Oy/+XL+HMCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"81672a97215eab68305195cf913124f63ae20f1024ac0d4dc806faab03ba377a","last_reissued_at":"2026-07-05T11:11:48.108014Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:11:48.108014Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognition","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bin Wang, Chong Zhu, Dacheng Tao, Dong Li, Fei Mi, Hailin Hu, Hanting Chen, Haoyu Wang, Jinpeng Li, Kai Han, Kaikai Song, Lin Li, Mingjian Zhu, Quan He, Wei He, Xinghao Chen, Xueyu Wu, Xu He, Yasheng Wang, Yehui Tang, Yifei Fu, Yu Luo, Yunhe Wang (and Other Contributors), Zhenni Bi","submitted_at":"2025-05-28T14:03:02Z","abstract_excerpt":"This work presents Pangu Embedded, an efficient Large Language Model (LLM) reasoner developed on Ascend Neural Processing Units (NPUs), featuring flexible fast and slow thinking capabilities. Pangu Embedded addresses the significant computational costs and inference latency challenges prevalent in existing reasoning-optimized LLMs. We propose a two-stage training framework for its construction. In Stage 1, the model is finetuned via an iterative distillation process, incorporating inter-iteration model merging to effectively aggregate complementary knowledge. This is followed by reinforcement "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.22375","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/2505.22375/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":"2505.22375","created_at":"2026-07-05T11:11:48.108073+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.22375v2","created_at":"2026-07-05T11:11:48.108073+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.22375","created_at":"2026-07-05T11:11:48.108073+00:00"},{"alias_kind":"pith_short_12","alias_value":"QFTSVFZBL2VW","created_at":"2026-07-05T11:11:48.108073+00:00"},{"alias_kind":"pith_short_16","alias_value":"QFTSVFZBL2VWQMCR","created_at":"2026-07-05T11:11:48.108073+00:00"},{"alias_kind":"pith_short_8","alias_value":"QFTSVFZB","created_at":"2026-07-05T11:11:48.108073+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07748","citing_title":"Selective Left-Shift: Turning Test-Time Compute and Difficulty-based Curation into Training Data for Low-Resource Code Generation","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2606.09061","citing_title":"Fairness-Aware and Latency-Controllable Scheduling for Chunked-Prefill LLM Serving","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30553","citing_title":"COSM: A Cooperative Scheduling Framework for Concurrent PIM and CPU Execution on Mobile Devices","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30553","citing_title":"COSM: A Cooperative Scheduling Framework for Concurrent PIM and CPU Execution on Mobile Devices","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26784","citing_title":"Ratio-Variance Regularized Policy Optimization","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28283","citing_title":"PrunePath: Towards Highly Structured Sparse Language Models","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09508","citing_title":"From Rigid to Dynamic: Entropy-Guided Adaptive Inference for Long-Context LLMs","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2601.17261","citing_title":"AGZO: Activation-Guided Zeroth-Order Optimization for LLM Fine-Tuning","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22104","citing_title":"OPERA: An Agent for Image Restoration with End-to-End Joint Planning-Execution Optimization","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2602.08324","citing_title":"Towards Efficient Large Language Reasoning Models via Extreme-Ratio Chain-of-Thought Compression","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20950","citing_title":"Focus-then-Context: Subject-Centric Progressive Visual Token Reduction for Vision-Language Models","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02702","citing_title":"TypePro: Boosting LLM-Based Type Inference via Inter-Procedural Slicing","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03644","citing_title":"AdapShot: Adaptive Many-Shot In-Context Learning with Semantic-Aware KV Cache Reuse","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05940","citing_title":"Near-Policy: Accelerating On-Policy Distillation via Asynchronous Generation and Selective Packing","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09752","citing_title":"A-IO: Adaptive Inference Orchestration for Memory-Bound NPUs","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14828","citing_title":"Pangu-ACE: Adaptive Cascaded Experts for Educational Response Generation on EduBench","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y","json":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y.json","graph_json":"https://pith.science/api/pith-number/QFTSVFZBL2VWQMCRSXHZCMJE6Y/graph.json","events_json":"https://pith.science/api/pith-number/QFTSVFZBL2VWQMCRSXHZCMJE6Y/events.json","paper":"https://pith.science/paper/QFTSVFZB"},"agent_actions":{"view_html":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y","download_json":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y.json","view_paper":"https://pith.science/paper/QFTSVFZB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.22375&json=true","fetch_graph":"https://pith.science/api/pith-number/QFTSVFZBL2VWQMCRSXHZCMJE6Y/graph.json","fetch_events":"https://pith.science/api/pith-number/QFTSVFZBL2VWQMCRSXHZCMJE6Y/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y/action/storage_attestation","attest_author":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y/action/author_attestation","sign_citation":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y/action/citation_signature","submit_replication":"https://pith.science/pith/QFTSVFZBL2VWQMCRSXHZCMJE6Y/action/replication_record"}},"created_at":"2026-07-05T11:11:48.108073+00:00","updated_at":"2026-07-05T11:11:48.108073+00:00"}