Diffusion large language models (dLLMs) present a promising alternative to dominant autoregressive models (ARMs) by the ability of parallel decoding at the expense of substantial computation and memory costs. Specifically, the cache mechanism for bidirectional attention in dLLMs demands large memory footprint, restricting their ability to handle long contexts under resource-limited settings. Existing cache eviction strategies are primarily designed for ARMs and fail to account for the role of mask tokens and specific characteristics in dLLMs, resulting in suboptimal performance.To address these challenges, we introduce MaskKV, a training-free cache eviction framework tailored to dLLMs, focusing on the effect of mask tokens in dLLMs. MaskKV is built on two key innovations: (1) a mask-query guided scoring mechanism that leverages attention weights to identify and evict less critical prompt tokens for each head; (2) an adaptive cache budgeting strategy that improves efficiency by reducing allocation in intermediate layers and concentrating resources on prompt-preferring heads. On LLaDA with MaskKV, compressing the KV cache to only 256 pairs (less than 5% of tokens) retains 94% of the full-cache performance on LongBench and achieves up to 31 × acceleration at 32k prompt length. Our code will be released on Github.

Internet memes emotion recognition is focused by many researchers. In this paper, we adopt BERT and ResNet for evaluation of detecting the emotions of Internet memes. We focus on solving the problem of data imbalance and data contains noise. We use RandAugment to enhance the data of the picture, and use Training Signal Annealing (TSA) to solve the impact of the imbalance of the label. At the same time, a new loss function is designed to ensure that the model is not affected by input noise which will improve the robustness of the model. We participated in sub-task a and our model based on BERT obtains 34.58% macro F1 score, ranking 10/32.

Offensive language has become pervasive in social media. In Offensive Language Identification tasks, it may be difficult to predict accurately only according to the surface words. So we try to dig deeper semantic information of text. This paper presents use an attention-based two layers bidirectional longshort memory neural network (BiLSTM) for semantic feature extraction. Additionally, a residual connection mechanism is used to synthesize two different deep features, and an emoji attention mechanism is used to extract semantic information of emojis in text. We participated in three sub-tasks of SemEval 2019 Task 6 as CN-HIT-MI.T team. Our macro-averaged F1-score in sub-task A is 0.768, ranking 28/103. We got 0.638 in sub-task B, ranking 30/75. In sub-task C, we got 0.549, ranking 22/65. We also tried some other methods of not submitting results.