Tag
The paper introduces Sharp Mode Connectivity (SMC) to enhance the robustness of weight-space interpolation in neural networks by optimizing for flatness along entire paths, achieving significant accuracy improvements under distribution shifts.
QwenMix-3.7 is a weight-interpolation merge of the Qwen3.6-27B and Qwen3.8-27B AI models, created using per-row norm preservation at t=0.5.
This paper presents ZooClaw-FashionSigLIP2, a fashion-specialized vision-language model that achieves superior retrieval performance through full fine-tuning with knowledge distillation and weight interpolation, outperforming larger backbones and LoRA. It also introduces a new high-quality benchmark, ZooClaw-Fashion, and analyzes structural biases in existing datasets.