Koziev/NLP_Datasets

Provides curated dialogue corpora (imageboard conversations, literary fiction, jokes, movie subtitles) with quality-scored JSONL annotations for relevance and specificity, plus synthetic Q&A datasets for arithmetic reasoning. Includes shallow-parsed sentence templates with morphologically-tagged noun phrase slots (~21M samples) and collocation/verb-phrase patterns extracted via automatic chunking, designed for training generative Russian chatbots compatible with ruGPT and sentence-BERT models. Datasets are distributed across Hugging Face Hub and GitHub with companion training scripts for fine-tuning dialogue systems and paraphrase detectors.

386 stars. No commits in the last 6 months.