Does your model actually use what it read? Synthetic long-context tests (1k-128k) with exact answers and guessing floors.
Krzysztof Sopyla
ksopyla
·
AI & ML interests
NLP, knowledge extraction, knowledge graphs, semantic similarity, model factfulness
Recent Activity
upvoted a collection about 20 hours ago
ncm updated a dataset 9 days ago
ksopyla/cogito-text-world updated a dataset 13 days ago
ksopyla/cogito-probe-propsOrganizations
models 8
ksopyla/gemma3-concepts-1b-c128-1bt
Text Generation • 1B • Updated • 23
ksopyla/concept-encoder-perceiver_mlm_H512L6C128_20260208_211633
Feature Extraction • Updated • 8
ksopyla/minipile-unigram-65k-50k
Updated
ksopyla/minipile-unigram-32k-50k
Updated
ksopyla/minipile-unigram-65k-100k
Updated
ksopyla/minipile-unigram-32k-100k
Updated
ksopyla/minipile-english-unigram-64k
Updated
ksopyla/minipile-english-unigram-32k
Updated