Rethinking Projector Training in Multimodal LLMs
cs.CV, cs.AI, cs.CL
Submitted: 2026-10-13
Updated: 2026-10-13
Comments: NeurIPS 2026 camera-ready
Code: https://github.com/mlbio-epfl/PORTAL
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
- AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Document Understanding
- SmolVLM: Redefining small and efficient multimodal models
- The Unseen Bias: How Norm Discrepancy in Pre-Norm MLLMs Leads to Visual Information Loss
- Sample-efficient Integration of New Modalities into Large Language Models
- Qwen3-VL Technical Report
- The Llama 3 Herd of Models
- Qwen2.5 Technical Report
- 2 OLMo 2 Furious
- Sigmoid Loss for Language Image Pre-Training
- MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models
- OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning
- MobileVLM : A Fast, Strong and Open Vision Language Assistant for Mobile Devices
- LatentLens: Revealing Highly Interpretable Visual Tokens in LLMs
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models