Tag: Multimodality

All the articles with the tag "Multimodality".

• Unraveling LoRA Interference: Orthogonal Subspaces for Robust Model Merging

  Published: 1 Jun, 2025 |  at  11:40 AM

  

  89.76 🤔

  本文提出OSRM方法，通过在微调前约束LoRA子空间以减少任务间干扰，显著提升了多个语言模型在八个GLUE数据集上的合并性能，同时保持单任务准确性。

• Shadow-FT: Tuning Instruct via Base

  Published: 25 May, 2025 |  at  11:25 AM

  

  89.60 🤔

  本文提出Shadow-FT框架，通过调优BASE模型并将权重更新直接移植到INSTRUCT模型，显著提升了大型语言模型在数学、编码和推理任务上的性能，同时不引入额外训练成本。

• Zebra-Llama: Towards Extremely Efficient Hybrid Models

  Published: 31 May, 2025 |  at  11:34 AM

  

  89.05 🤔

  Zebra-Llama通过结合状态空间模型和多头潜在注意力层，从预训练Transformer构建高效混合模型，显著降低KV缓存大小并提升推理吞吐量，同时保持或超越基线性能。

• An Efficient Sparse Kernel Generator for O(3)-Equivariant Deep Networks

  Published: 15 May, 2025 |  at  11:06 AM

  

  88.71 🤔

  This paper introduces a GPU sparse kernel generator for the Clebsch-Gordon tensor product in O(3)-equivariant deep networks, achieving significant speedups (up to 10x over e3nn and 1.3x-2.0x over cuEquivariance) by leveraging JIT compilation, static analysis, and kernel fusion, particularly enhancing performance in computational chemistry models like Nequip and MACE.

• Layer Swapping for Zero-Shot Cross-Lingual Transfer in Large Language Models

  Published: 31 May, 2025 |  at  11:35 AM

  

  88.25 🤔

  本文提出了一种层交换方法，通过将语言专家模型的顶部和底部层与数学专家模型的中间层重组，实现零样本跨语言迁移，在低资源语言的数学推理任务上显著提升性能达10%。