{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/revisiting-zeroth-order-optimization-for","title":"Revisiting Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning: A Benchmark","arxiv_id":"2402.11592","date":"2024-02-18","proceeding":null,"authors":["Yihua Zhang","Pingzhi Li","Junyuan Hong","Jiaxiang Li","Yimeng Zhang","Wenqing Zheng","Pin-Yu Chen","Jason D. Lee","Wotao Yin","Mingyi Hong","Zhangyang Wang","Sijia Liu","Tianlong Chen"],"abstract":"In the evolving landscape of natural language processing (NLP), fine-tuning pre-trained Large Language Models (LLMs) with first-order (FO) optimizers like SGD and Adam has become standard. Yet, as LLMs grow {in size}, the substantial memory overhead from back-propagation (BP) for FO gradient computation presents a significant challenge. Addressing this issue is crucial, especially for applications like on-device training where memory efficiency is paramount. This paper proposes a shift towards BP-free, zeroth-order (ZO) optimization as a solution for reducing memory costs during LLM fine-tuning, building on the initial concept introduced by MeZO. Unlike traditional ZO-SGD methods, our work expands the exploration to a wider array of ZO optimization techniques, through a comprehensive, first-of-its-kind benchmarking study across five LLM families (Roberta, OPT, LLaMA, Vicuna, Mistral), three task complexities, and five fine-tuning schemes. Our study unveils previously overlooked optimization principles, highlighting the importance of task alignment, the role of the forward gradient method, and the balance between algorithm complexity and fine-tuning performance. We further introduce novel enhancements to ZO optimization, including block-wise descent, hybrid training, and gradient sparsity. Our study offers a promising direction for achieving further memory-efficient LLM fine-tuning. Codes to reproduce all our experiments are at https://github.com/ZO-Bench/ZO-LLM .","url_abs":"https://arxiv.org/abs/2402.11592v3","url_pdf":"https://arxiv.org/pdf/2402.11592v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"revisiting-zeroth-order-optimization-for","repo_url":"https://github.com/zo-bench/zo-llm","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"GPL-3.0"}}],"tasks":[{"task_slug":"benchmarking","task_name":"Benchmarking"}],"methods":[{"method_slug":"adam","method_name":"Adam"},{"method_slug":"forward-gradient","method_name":"Forward gradient"},{"method_slug":"opt","method_name":"OPT"},{"method_slug":"sgd","method_name":"SGD"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2402.11592","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2402.11592"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/zo-bench/zo-llm","reach":{"status":"ok","spdx":"GPL-3.0"}}],"summary":{"ran_fixture":3,"ran":2,"ran_draft_wrong":3,"unverified":5},"by_repo_kind":{"official":{"samples":13,"ran":8,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":13,"samples":[{"code_sha256_prefix":"30d7eec482ebf6b1","entry":"repeat_kv","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/modeling_llama.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/modeling_llama.py","link_basis":"harvester_set","language":"python","status":"ran_fixture","verification_level":2,"contract_check":"RAISES","metamorphic_tier":"invariant","behaviour_fingerprint":true,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"30d7eec482ebf6b1"}},{"code_sha256_prefix":"d61c483a3c2b3156","entry":"apply_rotary_pos_emb","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/modeling_llama.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/modeling_llama.py","link_basis":"harvester_set","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"invariant","behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"d61c483a3c2b3156"}},{"code_sha256_prefix":"f725bc2d76076485","entry":"apply_rotary_pos_emb","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/modeling_mistral/modeling_mistral.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/modeling_mistral/modeling_mistral.py","link_basis":"harvester_set","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"invariant","behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"f725bc2d76076485"}},{"code_sha256_prefix":"f53e3216ab15d617","entry":"encode_prompt","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/utils.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/utils.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"f53e3216ab15d617"}},{"code_sha256_prefix":"b5adfd6ac99a1b18","entry":"find_module","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/lora.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/lora.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"b5adfd6ac99a1b18"}},{"code_sha256_prefix":"6a8d363117a347bd","entry":"forward_wrap_with_option_len","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/utils.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/utils.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"6a8d363117a347bd"}},{"code_sha256_prefix":"e7e75981cb464788","entry":"normalize_answer","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/metrics.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/metrics.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":true,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"e7e75981cb464788"}},{"code_sha256_prefix":"b99eea6376d1e212","entry":"rotate_half","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/modeling_llama.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/modeling_llama.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":"invariant","behaviour_fingerprint":true,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"b99eea6376d1e212"}},{"code_sha256_prefix":"bdb460cb247501dc","entry":"attn_forward_hook","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/prefix_tuning.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/prefix_tuning.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"bdb460cb247501dc"}},{"code_sha256_prefix":"84c658f2c9252878","entry":"calculate_metric","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/metrics.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/metrics.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"84c658f2c9252878"}},{"code_sha256_prefix":"b989d4bce26f77ce","entry":"f1","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/metrics.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/metrics.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"b989d4bce26f77ce"}},{"code_sha256_prefix":"f7c384f1e681ea7a","entry":"get_task","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/tasks.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/tasks.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"f7c384f1e681ea7a"}},{"code_sha256_prefix":"84920a8095dca4fa","entry":"prepare_inputs_for_generation","repo":"zo-bench/zo-llm","repo_kind":"official","path":"zo-bench/prefix_tuning.py","file_url":"https://github.com/zo-bench/zo-llm/blob/HEAD/zo-bench/prefix_tuning.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"GPL-3.0","inline_ok":false,"mcp_get_code":{"code_sha256":"84920a8095dca4fa"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}