Papers › Intrinsic Evaluation of Unlearning Using Parametric Knowledge Traces

Intrinsic Evaluation of Unlearning Using Parametric Knowledge Traces

17 Jun 2024arXiv:2406.11614archive 2025-07-28

Yihuai Hong, Lei Yu, Haiqin Yang, Shauli Ravfogel, Mor Geva

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.

PaperPDFCodeCode Syntology ran

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, on Syntology's MCP service (how to connect):

Code

Syntology 20 code samples harvested from 1 repository linked to this paper: 18 ran (of which 0 constructed an object rather than computing a result; 9 with no instrument failure: 4 honoured, 0 violated, 5 with no contract checked; 9 where Syntology's instrument failed) · 2 unverified. Where Syntology's instrument failed: 8 where our contract draft was wrong; 1 where the contract check could not run; failures of Syntology's instrument, not of the code.

By repository: official repository: 20 samples from 1 repository, 18 ran (9 with no instrument failure; 9 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

Levels written between 2026-09-22 and 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

yihuaihong/conceptvectors officialmentioned in papermentioned on GitHubpytorchCC-BY-4.0Syntology: 20 samples, 18 ran (9 instrument-failed) · 2 unverified report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

20 samples harvested: 18 ran (of which 0 constructed an object rather than computing a result; 9 with no instrument failure: 4 honoured, 0 violated, 5 with no contract checked; 9 where Syntology's instrument failed) · 2 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

4ran · honoured contract
8ran · our draft was wrong
1ran; the contract check could not run
5ran
2unverified

Licence: 20 of the 20 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from yihuaihong/conceptvectors. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. A line beginning “fixture:” under a sample that ran gives the arguments Syntology's fixture called the sample with, each as name:type/rank/element kind (? where none was recorded), and the type of what the fixture's run returned, exactly as the harness recorded them; the arguments and the returned type were recorded by two separate runs of the same fixture. No output value is shown, and a returned type says nothing about whether the output was right. “Level recorded” gives the UTC date when Syntology's graph wrote the sample's verification level, not when the code executed; where the level was written before Syntology's graph kept that date, the row says “level recorded, date not recorded”. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: 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. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

A function or class name links to its page when the name occurs in the harvested code of at least 5 papers; the name is a string, not an identity. A census tag appears when Syntology's census ran implementations with that name on one shared input per bucket: it gives how many were compared, from how many papers, and how many distinct outputs they produced, and names this sample's bucket when this sample was one of them.

yihuaihong/conceptvectors official repository 20 samples, 18 ran (9 instrument-failed) · 2 unverified
binary_log_probs in 30 papers memit/baselines/mend/losses.py ran · our draft was wrong CC-BY-4.0 · pointer only · 27659a9c234ffb11 · report fixture: signature pred:Tensor/2/float, targ:Tensor/2/int · returned dict · level recorded 2026-09-22
calculate_bleu in 10 papers evaluate_util.py ran · honoured contract fingerprinted CC-BY-4.0 · pointer only · 7f57090da94e6a92 · report fixture: signature reference:str/0/str, candidate:str/0/str · returned float · level recorded 2026-09-24
collate_fn in 500 papers data_module.py ran · honoured contract CC-BY-4.0 · pointer only · cdd171f20cd6f53d · report fixture: signature batch:list/1/tuple · returned tuple · level recorded 2026-09-24
convert_raw_data_to_model_format in 10 papers data_module.py ran CC-BY-4.0 · pointer only · d11b2101f4a1ec57 · report fixture: signature tokenizer:callable/0/?, max_length:int/0/int, text:str/0/str · returned tuple · level recorded 2026-09-24
custom_data_collator_forget in 6 papers dataloader.py ran; the contract check could not run (RAISES) CC-BY-4.0 · pointer only · 4e73f565fb762fa9 · report fixture: signature samples:list/1/tuple · returned list · level recorded 2026-09-24
evaluate in 514 papers evaluate_util.py ran CC-BY-4.0 · pointer only · b5f15de193c7c033 · report fixture: signature model:callable/0/?, tokenizer:callable/0/?, QA:list/1/str, text_completion:list/1/dict, unrelated_QA:list/1/str · returned tuple · level recorded 2026-09-24
evaluate in 514 papers Jailbreak/evaluate_util.py ran CC-BY-4.0 · pointer only · d06a2159d79e0268 · report fixture: signature model:callable/0/?, tokenizer:callable/0/?, QA:list/1/str, text_completion:list/1/dict, unrelated_QA:list/1/str · returned tuple · level recorded 2026-09-24
get_loss in 107 papers
census: 5 implementations from 5 papers, 5 distinct outputs across 4 buckets, one shared input per bucket
dataloader.py ran CC-BY-4.0 · pointer only · 650d5f6bfcd7d9bb · report fixture: signature output:Tensor/3/float, labels:Tensor/2/int · returned Tensor · level recorded 2026-09-24
get_total_len in 6 papers utils.py ran · honoured contract fingerprinted CC-BY-4.0 · pointer only · 6c542acea5aa211c · report fixture: signature name:str/0/str, forget_rate:str/0/str · returned int · level recorded 2026-09-24
hierarchical_subsequence in 47 papers memit/util/nethook.py ran CC-BY-4.0 · pointer only · 920b394e7ad80c53 · report fixture: signature sequential:callable/0/?, first:NoneType/0/?, last:NoneType/0/?, after:NoneType/0/?, upto:NoneType/0/?, share_weights:bool/0/bool, depth:int/0/int · returned Sequential · level recorded 2026-09-24
kl_loc_loss in 28 papers memit/baselines/mend/losses.py ran · our draft was wrong CC-BY-4.0 · pointer only · 8b02eb3a3733d653 · report fixture: signature pre:Tensor/3/float, post:Tensor/3/float, mask:Tensor/1/float · returned Tensor · level recorded 2026-09-24
merge_dicts in 14 papers utils.py ran · our draft was wrong CC-BY-4.0 · pointer only · 8076aab5782c4d02 · report fixture: signature a:dict/0/?, b:dict/0/? · returned dict · level recorded 2026-09-24
merge_pt_files evaluate_llama.py ran · our draft was wrong CC-BY-4.0 · pointer only · 366e41dac051e933 · report fixture: signature base_dir:str/0/str, num_files:int/0/int · returned list · level recorded 2026-09-24
merge_pt_files evaluate_olmo.py ran · honoured contract CC-BY-4.0 · pointer only · abc70a1f11fa71e2 · report fixture: signature base_dir:str/0/str, num_files:int/0/int · returned list · level recorded 2026-09-24
multiclass_log_probs in 20 papers memit/baselines/mend/losses.py ran · our draft was wrong CC-BY-4.0 · pointer only · 8ccc8886c554a1ce · report fixture: signature pred:Tensor/3/float, targ:Tensor/2/int, shift:bool/0/bool · returned dict · level recorded 2026-09-24
recursive_copy in 51 papers memit/util/nethook.py ran · our draft was wrong CC-BY-4.0 · pointer only · 70f6ab8bde55420e · report fixture: signature x:dict/0/?, clone:bool/0/bool, detach:bool/0/bool, retain_grad:bool/0/bool · returned dict · level recorded 2026-09-22
split_paragraph data_module.py ran · our draft was wrong fingerprinted CC-BY-4.0 · pointer only · 5b2b715b081b582e · report fixture: signature paragraph:str/0/str · returned list · level recorded 2026-09-24
subsequence in 51 papers memit/util/nethook.py ran · our draft was wrong CC-BY-4.0 · pointer only · 440ff98c2b1ae1aa · report fixture: signature sequential:callable/0/?, first_layer:NoneType/0/?, last_layer:NoneType/0/?, after_layer:NoneType/0/?, upto_layer:NoneType/0/?, single_layer:NoneType/0/?, share_weights:bool/0/bool · returned Sequential · level recorded 2026-09-24
evaluate in 514 papers forget.py unverified CC-BY-4.0 · pointer only · 0a20c1af0875a853 · report
get_model_identifiers_from_yaml in 12 papers utils.py unverified CC-BY-4.0 · pointer only · 06448b0513272463 · report

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

AttentionSoftmax

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections