Vision Language Models
PulseAugur coverage of Vision Language Models — every cluster mentioning Vision Language Models across labs, papers, and developer communities, ranked by signal.
6 天有情绪数据
-
ProcFunc library streamlines 3D generation and data creation in Python
A new Python library called ProcFunc has been developed for procedural 3D generation within Blender. This library offers a collection of user-friendly functions designed to simplify the creation, combination, and execut…
-
Stateful Transformers boost streaming inference; Intel releases AutoRound quantization toolkit
A new paper introduces a stateful transformer inference engine that significantly speeds up processing for streaming data by maintaining a persistent KV cache. This approach allows for query latency that is independent …
-
Apple researchers develop Direct Steering Optimization to mitigate AI bias
Researchers have developed Direct Steering Optimization (DSO), a novel method to mitigate bias in generative models like vision-language models (VLMs) and large language models (LLMs). DSO employs reinforcement learning…
-
VLMs struggle to interpret UI animations, new dataset reveals
Researchers have developed AniMINT, a new dataset comprising 300 annotated videos of UI animations, to evaluate how well Vision-Language Models (VLMs) understand dynamic interfaces. Current VLMs can detect basic motion …
-
New methods offer efficient data valuation for LLMs and VLMs
Two new research papers propose novel methods for data valuation in large language models (LLMs). The first, "For-Value," introduces an efficient forward-only framework that estimates data value using a single forward p…
-
GA2-CLIP paper introduces generic attribute anchors for VLM prompt tuning
Researchers have developed GA2-CLIP, a novel framework designed to enhance the generalization capabilities of Vision-Language Models (VLMs) in video tasks. This plug-and-play method addresses the issue of semantic space…
-
VLMs over-correct math OCR, hiding student errors; new metric PINK improves evaluation
Researchers have identified a significant issue in evaluating handwritten math OCR systems, particularly with Vision-Language Models (VLMs). These models often over-correct student errors instead of accurately transcrib…
-
New research explores GNN interpretability and multi-graph reasoning
Researchers are exploring new methods to enhance the interpretability and utility of Graph Neural Networks (GNNs). One paper investigates the critical role of node features in graph pooling, proposing that effective poo…
-
SMoES improves MoE-VLM efficiency and effectiveness with soft modality guidance
Researchers have introduced SMoES, a novel approach for guiding expert routing in Mixture-of-Experts (MoE) vision-language models (VLMs). This method utilizes dynamic soft modality scores to account for layer-dependent …
-
Hugging Face 推出用于高效 LLM 的先进量化技术
研究人员正在开发先进的量化技术，以提高大型语言模型 (LLM) 的效率。AutoRound、LATMiX 和 GSQ 等新方法旨在减小模型大小和计算需求，从而能够在功能较弱的硬件上进行部署。这些方法侧重于优化模型权重和激活在较低比特宽度下的表示方式，其中一些方法已达到与更高精度模型相当的准确性。创新包括用于训练后量化的新颖校准策略和用于提高鲁棒性的可学习仿射变换。