Trains a 125M-parameter Persian PLM on a curated 45GB corpus using vector semantic deduplication for domain balance, topping QA and NLI benchmarks while remaining competitive on NER and classification.
FarsTail: a Persian natural language inference dataset,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
IHUBERT: Vector-Based Semantic Deduplication and Domain-Balanced Pretraining for Persian Resources
Trains a 125M-parameter Persian PLM on a curated 45GB corpus using vector semantic deduplication for domain balance, topping QA and NLI benchmarks while remaining competitive on NER and classification.