Llama verbatim memorization probe (Meta)
Models get prompts sampled from their pretraining data. The score is the share of generations that reproduce the true continuation verbatim.
In-house Meta measurement on pretrained models, reported for several prompt and continuation lengths.
- Made by
- Meta
- Released
- July 2024
- Kind
- Refusing harmful requests
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- Kept private
- Last in a system card
- November 2024
- Last model added
- November 2024
3
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Llama verbatim memorization probe (Meta) yet.
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
3 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 23, 2024 |  Llama 3.1-8B technical report | Average verbatim memorization, pretrained, three scenarios with no overall (Table 24): English 50-gram 0.26%, All 50-gram 0.24%, All 1000-gram 1.11% | Safety | |
| Nov 23, 2024 | Llama 3.1-70B technical report | Average verbatim memorization, pretrained, three scenarios with no overall (Table 24): English 50-gram 0.60%, All 50-gram 0.55%, All 1000-gram 3.56% | Safety | |
| Nov 23, 2024 | Llama 3.1-405B technical report | Average verbatim memorization, pretrained, three scenarios with no overall (Table 24): English 50-gram 1.13%, All 50-gram 1.03%, All 1000-gram 3.91% | Safety |
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6% - Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts.
Gemini 4 Carbon 0% - DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics. Muse Spark 1.2 88.4%
- Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
- IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%