ItzSiden/TLumina_Pretrain_Dataset

# T-Lumina Pretrain Dataset ## 📊 Overview - Size: ~3.3GB - Total lines: ~5.6M - Language: - Bangla: ~89% - English: ~11% ## 🎯 Purpose This dataset is designed for pretraining a Bangla-focused educational language model (T-Lumina). ## 📚 Data Sources - Wikipedia-style articles - News text - Informational and educational content - Mixed Bangla-English contextual data ## 🧹 Processing - Cleaned (removed noise, refs, junk) - Filtered short/incomplete lines - Shuffled for better generalization ## ⚠️ Notes - Contains formal and semi-formal language - Not conversation-focused - Best used for base pretraining, followed by fine-tuning ## 🚀 Usage Suitable for: - Bangla LLM pretraining - Educational AI systems - Multilingual experiments --- license: mit ---