This paper introduces the approach of “Gradient Masters” for BLP-2025 Task 1: “Bangla Multitask Hate Speech Identification Shared Task”. We present an ensemble-based fine-tuning strategy for addressing subtasks 1A (hate-type classification) and 1B (target group classification) in YouTube comments. We propose a hybrid approach on a Bangla Language Model, which outperformed the baseline models and secured the 6th position in subtask 1A with a micro F1 score of 73.23% and the third position in subtask 1B with 73.28%. We conducted extensive experiments that evaluated the robustness of the model throughout the development and evaluation phases, including comparisons with other Language Model variants, to measure generalization in low-resource Bangla hate speech scenarios and data set coverage. In addition, we provide a detailed analysis of our findings, exploring misclassification patterns in the detection of hate speech.

This paper presents WoNBias, a curated Bengali dataset to identify gender-based biases, stereotypes, and harmful language directed at women. It merges digital sources- social media, blogs, news- with offline tactics comprising surveys and focus groups, alongside some existing corpora to compile a total of 31,484 entries (10,656 negative; 10,170 positive; 10,658 neutral). WoNBias reflects the sociocultural subtleties of bias in both Bengali digital and offline conversations. By bridging online and offline biased contexts, the dataset supports content moderation, policy interventions, and equitable NLP research for Bengali, a low-resource language critically underserved by existing tools. WoNBias aims to combat systemic gender discrimination against women on digital platforms, empowering researchers and practitioners to combat harmful narratives in Bengali-speaking communities.