Fine-tuned BanglaBERT v2 tokenizer for hate speech detection (Subtask 1B)

Fine-tuned BanglaBERT v2 for hate speech detection (Subtask 1B) - Accuracy: 0.7253

initial commit