{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:UNX26OZ7GEOI56JSFIEEPNMN42","short_pith_number":"pith:UNX26OZ7","schema_version":"1.0","canonical_sha256":"a36faf3b3f311c8ef9322a0847b58de691d38f0b614f0b127b68559e21c85362","source":{"kind":"arxiv","id":"2309.09055","version":1},"attestation_state":"computed","paper":{"title":"Exploring the impact of low-rank adaptation on the performance, efficiency, and regularization of RLHF","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dhawal Gupta, Mohit Iyyer, Simeng Sun","submitted_at":"2023-09-16T17:31:36Z","abstract_excerpt":"During the last stage of RLHF, a large language model is aligned to human intents via PPO training, a process that generally requires large-scale computational resources. In this technical report, we empirically investigate an efficient implementation of RLHF using low-rank adaptation (LoRA), which allows us to align the LLaMA 7B checkpoint on the Alpaca dataset using only two A100 GPUs instead of the eight required for full model fine-tuning. Despite tuning only 0.2% of LLaMA 7B's parameters, our implementation achieves better performance than the publicly-released AlpacaFarm checkpoint 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":"2309.09055","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-09-16T17:31:36Z","cross_cats_sorted":[],"title_canon_sha256":"f8c176a91e76cc17f8a20502b964a412e88401b7c45e0924860f19ed61646b04","abstract_canon_sha256":"ed79250ea9c6ce5b3963415540f2a0682e508b3a4f8a8b0f24ab8f377e92654f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:51:33.884371Z","signature_b64":"CYWuot7FvT0ZAkjYP2hhdxXJ6cLyQGGasYrKvSdkVYWD4yqWRXi8LeNJRDaCSIYvKfngu0JaZvgFc1CxaPguDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a36faf3b3f311c8ef9322a0847b58de691d38f0b614f0b127b68559e21c85362","last_reissued_at":"2026-07-05T06:51:33.883886Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:51:33.883886Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Exploring the impact of low-rank adaptation on the performance, efficiency, and regularization of RLHF","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dhawal Gupta, Mohit Iyyer, Simeng Sun","submitted_at":"2023-09-16T17:31:36Z","abstract_excerpt":"During the last stage of RLHF, a large language model is aligned to human intents via PPO training, a process that generally requires large-scale computational resources. In this technical report, we empirically investigate an efficient implementation of RLHF using low-rank adaptation (LoRA), which allows us to align the LLaMA 7B checkpoint on the Alpaca dataset using only two A100 GPUs instead of the eight required for full model fine-tuning. Despite tuning only 0.2% of LLaMA 7B's parameters, our implementation achieves better performance than the publicly-released AlpacaFarm checkpoint with "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2309.09055","kind":"arxiv","version":1},"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/2309.09055/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":"2309.09055","created_at":"2026-07-05T06:51:33.883935+00:00"},{"alias_kind":"arxiv_version","alias_value":"2309.09055v1","created_at":"2026-07-05T06:51:33.883935+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2309.09055","created_at":"2026-07-05T06:51:33.883935+00:00"},{"alias_kind":"pith_short_12","alias_value":"UNX26OZ7GEOI","created_at":"2026-07-05T06:51:33.883935+00:00"},{"alias_kind":"pith_short_16","alias_value":"UNX26OZ7GEOI56JS","created_at":"2026-07-05T06:51:33.883935+00:00"},{"alias_kind":"pith_short_8","alias_value":"UNX26OZ7","created_at":"2026-07-05T06:51:33.883935+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.07111","citing_title":"Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21905","citing_title":"Low-Rank Adaptation Redux for Large Models","ref_index":180,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07111","citing_title":"Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42","json":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42.json","graph_json":"https://pith.science/api/pith-number/UNX26OZ7GEOI56JSFIEEPNMN42/graph.json","events_json":"https://pith.science/api/pith-number/UNX26OZ7GEOI56JSFIEEPNMN42/events.json","paper":"https://pith.science/paper/UNX26OZ7"},"agent_actions":{"view_html":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42","download_json":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42.json","view_paper":"https://pith.science/paper/UNX26OZ7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2309.09055&json=true","fetch_graph":"https://pith.science/api/pith-number/UNX26OZ7GEOI56JSFIEEPNMN42/graph.json","fetch_events":"https://pith.science/api/pith-number/UNX26OZ7GEOI56JSFIEEPNMN42/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42/action/storage_attestation","attest_author":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42/action/author_attestation","sign_citation":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42/action/citation_signature","submit_replication":"https://pith.science/pith/UNX26OZ7GEOI56JSFIEEPNMN42/action/replication_record"}},"created_at":"2026-07-05T06:51:33.883935+00:00","updated_at":"2026-07-05T06:51:33.883935+00:00"}