{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:O77O4AGR4UTLVF57SM5MY2LZOE","short_pith_number":"pith:O77O4AGR","schema_version":"1.0","canonical_sha256":"77feee00d1e526ba97bf933acc6979713bbd4cd65f9c776eb8d808e9a2b65cd5","source":{"kind":"arxiv","id":"2604.01193","version":2},"attestation_state":"computed","paper":{"title":"Embarrassingly Simple Self-Distillation Improves Code Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Huangjie Zheng, Navdeep Jaitly, Richard He Bai, Ronan Collobert, Ruixiang Zhang, Yizhe Zhang","submitted_at":"2026-04-01T17:39:50Z","abstract_excerpt":"Can a large language model (LLM) improve at code generation using only its own raw outputs, without a verifier, a teacher model, or reinforcement learning? We answer in the affirmative with simple self-distillation (SSD): sample solutions from the model with certain temperature and truncation configurations, then fine-tune on those samples with standard supervised fine-tuning. SSD improves Qwen3-30B-Instruct from 42.4% to 55.3% pass@1 on LiveCodeBench v6, with gains concentrating on harder problems, and it generalizes across Qwen and Llama models at 4B, 8B, and 30B scale, including both instru"},"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":"2604.01193","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-04-01T17:39:50Z","cross_cats_sorted":[],"title_canon_sha256":"e81295bd64aa675fea672b88583930afd3edf9a1b0e1823ff7b204f2a0a11270","abstract_canon_sha256":"52583ec31c0baa00f1dc1273a9f49d642ea5b027c2d8f8ef2c1c2c54c9897db0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T00:15:25.484752Z","signature_b64":"0V7nYq437/LKFxvsu+dVqid1nz4SlYeCvHyNCXVLTNXg8Ad1U64Rg267rjdw5fpUlB8ltQCZZNTy5IbyUAAUAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"77feee00d1e526ba97bf933acc6979713bbd4cd65f9c776eb8d808e9a2b65cd5","last_reissued_at":"2026-06-26T00:15:25.484008Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T00:15:25.484008Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Embarrassingly Simple Self-Distillation Improves Code Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Huangjie Zheng, Navdeep Jaitly, Richard He Bai, Ronan Collobert, Ruixiang Zhang, Yizhe Zhang","submitted_at":"2026-04-01T17:39:50Z","abstract_excerpt":"Can a large language model (LLM) improve at code generation using only its own raw outputs, without a verifier, a teacher model, or reinforcement learning? We answer in the affirmative with simple self-distillation (SSD): sample solutions from the model with certain temperature and truncation configurations, then fine-tune on those samples with standard supervised fine-tuning. SSD improves Qwen3-30B-Instruct from 42.4% to 55.3% pass@1 on LiveCodeBench v6, with gains concentrating on harder problems, and it generalizes across Qwen and Llama models at 4B, 8B, and 30B scale, including both instru"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.01193","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/2604.01193/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":"2604.01193","created_at":"2026-06-26T00:15:25.484094+00:00"},{"alias_kind":"arxiv_version","alias_value":"2604.01193v2","created_at":"2026-06-26T00:15:25.484094+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.01193","created_at":"2026-06-26T00:15:25.484094+00:00"},{"alias_kind":"pith_short_12","alias_value":"O77O4AGR4UTL","created_at":"2026-06-26T00:15:25.484094+00:00"},{"alias_kind":"pith_short_16","alias_value":"O77O4AGR4UTLVF57","created_at":"2026-06-26T00:15:25.484094+00:00"},{"alias_kind":"pith_short_8","alias_value":"O77O4AGR","created_at":"2026-06-26T00:15:25.484094+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":14,"sample":[{"citing_arxiv_id":"2605.18141","citing_title":"A Brief Overview: On-Policy Self-Distillation In Large Language Models","ref_index":56,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06597","citing_title":"UniSD: Towards a Unified Self-Distillation Framework for Large Language Models","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.22263","citing_title":"Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning","ref_index":60,"is_internal_anchor":true},{"citing_arxiv_id":"2605.22675","citing_title":"Self-Policy Distillation via Capability-Selective Subspace Projection","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18529","citing_title":"AMR-SD: Asymmetric Meta-Reflective Self-Distillation for Token-Level Credit Assignment","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18141","citing_title":"A Brief Overview: On-Policy Self-Distillation In Large Language Models","ref_index":56,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13255","citing_title":"Respecting Self-Uncertainty in On-Policy Self-Distillation for Efficient LLM Reasoning","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2604.24809","citing_title":"Nautile-370M: Spectral Memory Meets Attention in a Small Reasoning Model","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06597","citing_title":"UniSD: Towards a Unified Self-Distillation Framework for Large Language Models","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05040","citing_title":"Preference-Based Self-Distillation: Beyond KL Matching via Reward Regularization","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.04542","citing_title":"Power Distribution Bridges Sampling, Self-Reward RL, and Self-Distillation","ref_index":138,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01130","citing_title":"Iterative Finetuning is Mostly Idempotent","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08532","citing_title":"Self-Improving 4D Perception via Self-Distillation","ref_index":77,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05851","citing_title":"Hypothesis generation and updating in large language models","ref_index":65,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE","json":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE.json","graph_json":"https://pith.science/api/pith-number/O77O4AGR4UTLVF57SM5MY2LZOE/graph.json","events_json":"https://pith.science/api/pith-number/O77O4AGR4UTLVF57SM5MY2LZOE/events.json","paper":"https://pith.science/paper/O77O4AGR"},"agent_actions":{"view_html":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE","download_json":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE.json","view_paper":"https://pith.science/paper/O77O4AGR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2604.01193&json=true","fetch_graph":"https://pith.science/api/pith-number/O77O4AGR4UTLVF57SM5MY2LZOE/graph.json","fetch_events":"https://pith.science/api/pith-number/O77O4AGR4UTLVF57SM5MY2LZOE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE/action/storage_attestation","attest_author":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE/action/author_attestation","sign_citation":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE/action/citation_signature","submit_replication":"https://pith.science/pith/O77O4AGR4UTLVF57SM5MY2LZOE/action/replication_record"}},"created_at":"2026-06-26T00:15:25.484094+00:00","updated_at":"2026-06-26T00:15:25.484094+00:00"}