{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PCZDFVJWUD6MNTBQZQNOEGGZHZ","short_pith_number":"pith:PCZDFVJW","schema_version":"1.0","canonical_sha256":"78b232d536a0fcc6cc30cc1ae218d93e76125ffc8f00d9f64acddda3ea8919a6","source":{"kind":"arxiv","id":"2410.10989","version":3},"attestation_state":"computed","paper":{"title":"Liger Kernel: Efficient Triton Kernels for LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.DC"],"primary_cat":"cs.LG","authors_text":"Haowen Ning, Pin-Lun Hsu, Qingquan Song, Shao Tang, Shivam Sahni, Siyu Zhu, Steven Shimizu, Vignesh Kothapalli, Yanning Chen, Yun Dai","submitted_at":"2024-10-14T18:17:01Z","abstract_excerpt":"Training Large Language Models (LLMs) efficiently at scale presents a formidable challenge, driven by their ever-increasing computational demands and the need for enhanced performance. In this work, we introduce Liger-Kernel, an open-sourced set of Triton kernels developed specifically for LLM training. With kernel optimization techniques like kernel operation fusing and input chunking, our kernels achieve on average a 20% increase in training throughput and a 60% reduction in GPU memory usage for popular LLMs compared to HuggingFace implementations. In addition, Liger-Kernel is designed with "},"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":"2410.10989","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-14T18:17:01Z","cross_cats_sorted":["cs.AI","cs.CL","cs.DC"],"title_canon_sha256":"c2fd0ee05280510eac0a6988dafe9ef1b3afc3176391a6c479c0462271b62bda","abstract_canon_sha256":"aae4146ddb9f9c2dc370a0e62b0cc22eee09f0030735c0c0c4aeb1d027cae0d7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:04:40.278272Z","signature_b64":"/CDX5qIHVr+JpX3YVS+PXNu1m3rWuikAOBW7dmTizCuZjIbGfxumdfnRgPFvGJEzPIv4z5T0gbPfEtltvsUsBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"78b232d536a0fcc6cc30cc1ae218d93e76125ffc8f00d9f64acddda3ea8919a6","last_reissued_at":"2026-07-05T10:04:40.277646Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:04:40.277646Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Liger Kernel: Efficient Triton Kernels for LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.DC"],"primary_cat":"cs.LG","authors_text":"Haowen Ning, Pin-Lun Hsu, Qingquan Song, Shao Tang, Shivam Sahni, Siyu Zhu, Steven Shimizu, Vignesh Kothapalli, Yanning Chen, Yun Dai","submitted_at":"2024-10-14T18:17:01Z","abstract_excerpt":"Training Large Language Models (LLMs) efficiently at scale presents a formidable challenge, driven by their ever-increasing computational demands and the need for enhanced performance. In this work, we introduce Liger-Kernel, an open-sourced set of Triton kernels developed specifically for LLM training. With kernel optimization techniques like kernel operation fusing and input chunking, our kernels achieve on average a 20% increase in training throughput and a 60% reduction in GPU memory usage for popular LLMs compared to HuggingFace implementations. In addition, Liger-Kernel is designed with "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.10989","kind":"arxiv","version":3},"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/2410.10989/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":"2410.10989","created_at":"2026-07-05T10:04:40.277742+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.10989v3","created_at":"2026-07-05T10:04:40.277742+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.10989","created_at":"2026-07-05T10:04:40.277742+00:00"},{"alias_kind":"pith_short_12","alias_value":"PCZDFVJWUD6M","created_at":"2026-07-05T10:04:40.277742+00:00"},{"alias_kind":"pith_short_16","alias_value":"PCZDFVJWUD6MNTBQ","created_at":"2026-07-05T10:04:40.277742+00:00"},{"alias_kind":"pith_short_8","alias_value":"PCZDFVJW","created_at":"2026-07-05T10:04:40.277742+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24937","citing_title":"The Hitchhiker's Guide to Agentic AI: From Foundations to Systems","ref_index":98,"is_internal_anchor":false},{"citing_arxiv_id":"2606.13578","citing_title":"LabVLA: Grounding Vision-Language-Action Models in Scientific Laboratories","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28593","citing_title":"Animation2Code: Evaluating Temporal Visual Reasoning in Video-to-Code Generation","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14790","citing_title":"Graphs of Research: Citation Evolution Graphs as Supervision for Research Idea Generation","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15913","citing_title":"Towards Generalization of Block Attention via Automatic Segmentation and Block Distillation","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2412.04468","citing_title":"NVILA: Efficient Frontier Visual Language Models","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2502.17421","citing_title":"LongSpec: Long-Context Lossless Speculative Decoding with Efficient Drafting and Verification","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15913","citing_title":"Towards Generalization of Block Attention via Automatic Segmentation and Block Distillation","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2509.09682","citing_title":"Faster and Memory-Efficient Training of Sequential Recommendation Models for Large Catalogs","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2509.12266","citing_title":"Genome-Factory: A Library for Tuning, Deploying, and Interpreting Genomic Foundation Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2602.08819","citing_title":"Bayesian Preference Learning for Test-Time Steerable Reward Models","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19269","citing_title":"CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15913","citing_title":"Towards Generalization of Block Attention via Automatic Segmentation and Block Distillation","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19269","citing_title":"CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16826","citing_title":"Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05942","citing_title":"BOSCH: Black-Box Binary Optimization for Short-Context Attention-Head Selection in LLMs","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ","json":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ.json","graph_json":"https://pith.science/api/pith-number/PCZDFVJWUD6MNTBQZQNOEGGZHZ/graph.json","events_json":"https://pith.science/api/pith-number/PCZDFVJWUD6MNTBQZQNOEGGZHZ/events.json","paper":"https://pith.science/paper/PCZDFVJW"},"agent_actions":{"view_html":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ","download_json":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ.json","view_paper":"https://pith.science/paper/PCZDFVJW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.10989&json=true","fetch_graph":"https://pith.science/api/pith-number/PCZDFVJWUD6MNTBQZQNOEGGZHZ/graph.json","fetch_events":"https://pith.science/api/pith-number/PCZDFVJWUD6MNTBQZQNOEGGZHZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ/action/storage_attestation","attest_author":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ/action/author_attestation","sign_citation":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ/action/citation_signature","submit_replication":"https://pith.science/pith/PCZDFVJWUD6MNTBQZQNOEGGZHZ/action/replication_record"}},"created_at":"2026-07-05T10:04:40.277742+00:00","updated_at":"2026-07-05T10:04:40.277742+00:00"}