{"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/intrinsic-evaluation-of-unlearning-using","title":"Intrinsic Evaluation of Unlearning Using Parametric Knowledge Traces","arxiv_id":"2406.11614","date":"2024-06-17","proceeding":null,"authors":["Yihuai Hong","Lei Yu","Haiqin Yang","Shauli Ravfogel","Mor Geva"],"abstract":"The task of \"unlearning\" certain concepts in large language models (LLMs) has attracted immense attention recently, due to its importance in mitigating undesirable model behaviours, such as the generation of harmful, private, or incorrect information. Current protocols to evaluate unlearning methods largely rely on behavioral tests, without monitoring the presence of unlearned knowledge within the model's parameters. This residual knowledge can be adversarially exploited to recover the erased information post-unlearning. We argue that unlearning should also be evaluated internally, by considering changes in the parametric knowledge traces of the unlearned concepts. To this end, we propose a general evaluation methodology that leverages vocabulary projections to inspect concepts encoded in model parameters. We use this approach to localize \"concept vectors\" - parameter vectors that encode concrete concepts - and construct ConceptVectors, a benchmark dataset containing hundreds of common concepts and their parametric knowledge traces within two open-source LLMs. Evaluation on ConceptVectors shows that existing unlearning methods minimally impact concept vectors and mostly suppress them during inference, while directly ablating these vectors demonstrably removes the associated knowledge and significantly reduces the model's susceptibility to adversarial manipulation. Our results highlight limitations in behavioral-based unlearning evaluations and call for future work to include parameter-based evaluations. To support this, we release our code and benchmark at https://github.com/yihuaihong/ConceptVectors.","url_abs":"https://arxiv.org/abs/2406.11614v2","url_pdf":"https://arxiv.org/pdf/2406.11614v2.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":"intrinsic-evaluation-of-unlearning-using","repo_url":"https://github.com/yihuaihong/conceptvectors","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"CC-BY-4.0"}}],"tasks":[],"methods":[{"method_slug":"attention","method_name":"Attention"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2406.11614","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2406.11614"}},"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/yihuaihong/conceptvectors","reach":{"status":"ok","spdx":"CC-BY-4.0"}}],"summary":{"ran_draft_wrong":2,"ran":16,"unverified":2},"by_repo_kind":{"official":{"samples":20,"ran":18,"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":20,"samples":[{"code_sha256_prefix":"27659a9c234ffb11","entry":"binary_log_probs","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"memit/baselines/mend/losses.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/memit/baselines/mend/losses.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":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"27659a9c234ffb11"}},{"code_sha256_prefix":"7f57090da94e6a92","entry":"calculate_bleu","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"evaluate_util.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/evaluate_util.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"7f57090da94e6a92"}},{"code_sha256_prefix":"cdd171f20cd6f53d","entry":"collate_fn","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"data_module.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/data_module.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"cdd171f20cd6f53d"}},{"code_sha256_prefix":"d11b2101f4a1ec57","entry":"convert_raw_data_to_model_format","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"data_module.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/data_module.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"d11b2101f4a1ec57"}},{"code_sha256_prefix":"4e73f565fb762fa9","entry":"custom_data_collator_forget","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"dataloader.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/dataloader.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"4e73f565fb762fa9"}},{"code_sha256_prefix":"b5f15de193c7c033","entry":"evaluate","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"evaluate_util.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/evaluate_util.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"b5f15de193c7c033"}},{"code_sha256_prefix":"d06a2159d79e0268","entry":"evaluate","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"Jailbreak/evaluate_util.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/Jailbreak/evaluate_util.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"d06a2159d79e0268"}},{"code_sha256_prefix":"650d5f6bfcd7d9bb","entry":"get_loss","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"dataloader.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/dataloader.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"650d5f6bfcd7d9bb"}},{"code_sha256_prefix":"6c542acea5aa211c","entry":"get_total_len","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"utils.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/utils.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"6c542acea5aa211c"}},{"code_sha256_prefix":"920b394e7ad80c53","entry":"hierarchical_subsequence","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"memit/util/nethook.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/memit/util/nethook.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"920b394e7ad80c53"}},{"code_sha256_prefix":"8b02eb3a3733d653","entry":"kl_loc_loss","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"memit/baselines/mend/losses.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/memit/baselines/mend/losses.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"8b02eb3a3733d653"}},{"code_sha256_prefix":"8076aab5782c4d02","entry":"merge_dicts","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"utils.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/utils.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"8076aab5782c4d02"}},{"code_sha256_prefix":"366e41dac051e933","entry":"merge_pt_files","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"evaluate_llama.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/evaluate_llama.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"366e41dac051e933"}},{"code_sha256_prefix":"abc70a1f11fa71e2","entry":"merge_pt_files","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"evaluate_olmo.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/evaluate_olmo.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"abc70a1f11fa71e2"}},{"code_sha256_prefix":"8ccc8886c554a1ce","entry":"multiclass_log_probs","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"memit/baselines/mend/losses.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/memit/baselines/mend/losses.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"8ccc8886c554a1ce"}},{"code_sha256_prefix":"70f6ab8bde55420e","entry":"recursive_copy","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"memit/util/nethook.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/memit/util/nethook.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"70f6ab8bde55420e"}},{"code_sha256_prefix":"5b2b715b081b582e","entry":"split_paragraph","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"data_module.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/data_module.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"5b2b715b081b582e"}},{"code_sha256_prefix":"440ff98c2b1ae1aa","entry":"subsequence","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"memit/util/nethook.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/memit/util/nethook.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"440ff98c2b1ae1aa"}},{"code_sha256_prefix":"0a20c1af0875a853","entry":"evaluate","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"forget.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/forget.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"0a20c1af0875a853"}},{"code_sha256_prefix":"06448b0513272463","entry":"get_model_identifiers_from_yaml","repo":"yihuaihong/conceptvectors","repo_kind":"official","path":"utils.py","file_url":"https://github.com/yihuaihong/conceptvectors/blob/HEAD/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC-BY-4.0","inline_ok":false,"mcp_get_code":{"code_sha256":"06448b0513272463"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}