datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
shadow-llm-mia-signals
Shadow LLM MIA Signals (OLMo-2-1B)
Membership Inference Attack (MIA) signal tensors extracted from 128 shadow models
fine-tuned from allenai/OLMo-2-0425-1B.
Overview
This dataset enables research on membership inference attacks against large language models.
Each of 128 shadow models was trained on a different random subset of 64 out of 128 candidate
documents from the OLMo-mix-1124 pretraining dataset.
For each (model, document) pair, we extracted softmax prediction… See the full description on the dataset page: https://huggingface.co/datasets/matthewwicker/shadow-llm-mia-signals.e15-context-budget
SignalDepth E15 Context Budget
This is a small prompt-sensitivity benchmark slice for separating two explanations that often get conflated:
the prompt is too short
the task contract is underspecified
The narrow result: on this deterministic Python code-task suite, making sparse prompts longer did not help. Making the task contract explicit did.
Key Result
Condition
Average pass rate
Read
short_sparse
0.25
short and underspecified
long_sparse
0.25
longer… See the full description on the dataset page: https://huggingface.co/datasets/signaldepth/e15-context-budget.