Benchmarking Post-Training Quantization of Large Language Models under Microscaling Floating Point Formats

Microscaling Floating-Point (MXFP) has emerged as a promising low-precision format for large language models (LLMs). Despite various post-training quantization (PTQ) algorithms being proposed, they mostly focus on integer quantization, while their applicability and behavior under MXFP formats remain largely unexplored. To address this gap, this work conducts a systematic investigation of PTQ under MXFP formats, encompassing over 7 PTQ algorithms, 15 evaluation benchmarks, and 3 LLM families. The key findings include: 1) MXFP8 consistently achieves near-lossless performance, while MXFP4 introduces substantial accuracy degradation and remains challenging; 2) PTQ effectiveness under MXFP depends strongly on format compatibility, with some algorithmic paradigms being consistently more effective than others; 3) PTQ performance exhibits highly consistent trends across model families and modalities, in particular, quantization sensitivity is dominated by the language model rather than the vision encoder in multimodal LLMs; 4) The scaling factor of quantization is a critical error source in MXFP4, and a simple pre-scale optimization strategy can significantly mitigate its impact. Together, these results provide practical guidance on adapting existing PTQ methods to MXFP quantization.

Robust Divergence Learning for Missing-Modality Segmentation

Multimodal Magnetic Resonance Imaging (MRI) provides essential complementary information for analyzing brain tumor subregions. While methods using four common MRI modalities for automatic segmentation have shown success, they often face challenges with missing modalities due to image quality issues, inconsistent protocols, allergic reactions, or cost factors. Thus, developing a segmentation paradigm that handles missing modalities is clinically valuable. A novel single-modality parallel processing network framework based on H\"older divergence and mutual information is introduced. Each modality is independently input into a shared network backbone for parallel processing, preserving unique information. Additionally, a dynamic sharing framework is introduced that adjusts network parameters based on modality availability. A H\"older divergence and mutual information-based loss functions are used for evaluating discrepancies between predictions and labels. Extensive testing on the BraTS 2018 and BraTS 2020 datasets demonstrates that our method outperforms existing techniques in handling missing modalities and validates each component's effectiveness.

Unveiling Incomplete Modality Brain Tumor Segmentation: Leveraging Masked Predicted Auto-Encoder and Divergence Learning

Brain tumor segmentation remains a significant challenge, particularly in the context of multi-modal magnetic resonance imaging (MRI) where missing modality images are common in clinical settings, leading to reduced segmentation accuracy. To address this issue, we propose a novel strategy, which is called masked predicted pre-training, enabling robust feature learning from incomplete modality data. Additionally, in the fine-tuning phase, we utilize a knowledge distillation technique to align features between complete and missing modality data, simultaneously enhancing model robustness. Notably, we leverage the Holder pseudo-divergence instead of the KLD for distillation loss, offering improve mathematical interpretability and properties. Extensive experiments on the BRATS2018 and BRATS2020 datasets demonstrate significant performance enhancements compared to existing state-of-the-art methods.