AlexWortega/openjev-data

openjev training data — everything except the benchmark part The mixture behind AlexWortega/openjev: a Qwen3.5 decoder trained as a 3-way NLI cross-encoder, where every downstream use — reranking, fact-checking, typed decisions, agent policies — is "each option becomes a hypothesis, score = P(entailment)". 2,346,729 rows. One row is one (premise, hypothesis) pair with a label. from datasets import load_dataset ds = load_dataset("AlexWortega/openjev-data"… See the full description on the dataset page: huggingface.co/datasets/AlexWortega/openjev-data.