# Secret Data Extraction on 100 Targeted Secrets OKVQA DocVQA VQAv2 Source: [Secret Data Extraction on 100 Targeted Secrets OKVQA DocVQA VQAv2]() Collection index: [llms.txt]() Data updated: 2026-06-17T06:31:06.324Z Evaluation of Secret Data Extraction performance on the combined dataset comprising 100 Targeted Secrets, OKVQA, DocVQA, and VQAv2 splits/subsets. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Secret Data Extraction | | Dataset (including any stated split) | 100 Targeted Secrets OKVQA DocVQA VQAv2 | | Metric | P(s\|x) \> 0.10 Rate | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 18 of 18 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [P(s\|x) \> 0.10 Rate](): Higher is better; unit: PERCENT. - [P(s\|x) \> 0.50 Rate](): Higher is better; unit: PERCENT. ## Recorded results | Method | P(s\|x) \> 0.10 Rate | Result date | Report paper | | --- | --- | --- | --- | | DistilGPT-2 (82M) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | GPT-2 Small (124M) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | GPT-Neo (125M) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | Pythia (160M) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | OPT (250M) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | GPT-2 Medium (355M) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | Llama2 (7B) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | Llama2 (13B) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | Llama3.2 (1B) | 100 | Not recorded | 6a322a20d33dfb23e783eb07 | | GPT-2 Medium (355M) | 15 | Not recorded | 6a322a20d33dfb23e783eb07 | | OPT (250M) | 10 | Not recorded | 6a322a20d33dfb23e783eb07 | | Pythia (160M) | 6 | Not recorded | 6a322a20d33dfb23e783eb07 | | GPT-2 Small (124M) | 3 | Not recorded | 6a322a20d33dfb23e783eb07 | | DistilGPT-2 (82M) | 2 | Not recorded | 6a322a20d33dfb23e783eb07 | | GPT-Neo (125M) | 2 | Not recorded | 6a322a20d33dfb23e783eb07 | | Llama2 (7B) | 0 | Not recorded | 6a322a20d33dfb23e783eb07 | | Llama2 (13B) | 0 | Not recorded | 6a322a20d33dfb23e783eb07 | | Llama3.2 (1B) | 0 | Not recorded | 6a322a20d33dfb23e783eb07 | ### DistilGPT-2 (82M) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02bc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning"}. This record does not assert independent reproduction. ### GPT-2 Small (124M) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02be. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning"}. This record does not assert independent reproduction. ### GPT-Neo (125M) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02c0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning"}. This record does not assert independent reproduction. ### Pythia (160M) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02c2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning"}. This record does not assert independent reproduction. ### OPT (250M) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02c4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning"}. This record does not assert independent reproduction. ### GPT-2 Medium (355M) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02c6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning"}. This record does not assert independent reproduction. ### Llama2 (7B) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02c8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning","LoRA fine-tuning":"true"}. This record does not assert independent reproduction. ### Llama2 (13B) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02ca. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning","LoRA fine-tuning":"true"}. This record does not assert independent reproduction. ### Llama3.2 (1B) P(s\|x) \> 0.10 Rate: 100. Result ID: 6a323db5fbc373592c2f02cc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"After poisoning","LoRA fine-tuning":"true"}. This record does not assert independent reproduction. ### GPT-2 Medium (355M) P(s\|x) \> 0.10 Rate: 15. Result ID: 6a323db5fbc373592c2f02c5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting"}. This record does not assert independent reproduction. ### OPT (250M) P(s\|x) \> 0.10 Rate: 10. Result ID: 6a323db5fbc373592c2f02c3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting"}. This record does not assert independent reproduction. ### Pythia (160M) P(s\|x) \> 0.10 Rate: 6. Result ID: 6a323db5fbc373592c2f02c1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting"}. This record does not assert independent reproduction. ### GPT-2 Small (124M) P(s\|x) \> 0.10 Rate: 3. Result ID: 6a323db5fbc373592c2f02bd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting"}. This record does not assert independent reproduction. ### DistilGPT-2 (82M) P(s\|x) \> 0.10 Rate: 2. Result ID: 6a323db5fbc373592c2f02bb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting"}. This record does not assert independent reproduction. ### GPT-Neo (125M) P(s\|x) \> 0.10 Rate: 2. Result ID: 6a323db5fbc373592c2f02bf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting"}. This record does not assert independent reproduction. ### Llama2 (7B) P(s\|x) \> 0.10 Rate: 0. Result ID: 6a323db5fbc373592c2f02c7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting","LoRA fine-tuning":"true"}. This record does not assert independent reproduction. ### Llama2 (13B) P(s\|x) \> 0.10 Rate: 0. Result ID: 6a323db5fbc373592c2f02c9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting","LoRA fine-tuning":"true"}. This record does not assert independent reproduction. ### Llama3.2 (1B) P(s\|x) \> 0.10 Rate: 0. Result ID: 6a323db5fbc373592c2f02cb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"Baseline setting","LoRA fine-tuning":"true"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.