Engineering
BAGEL 复现
ByteDance's BAGEL model uses a Mixture-of-Transformer-Experts architecture for unified image understanding, generation, and editing, requiring Ampere GPUs for full bfloat16 support.
Recently
Though the road be long, I will seek without end.
Engineering
VILA-U 复现
A technical guide for reproducing the VILA-U multimodal model on AutoDL, covering environment setup, storage optimization, model download, inference, and common troubleshooting.
Regularization & Optimization
Regularization prevents overfitting by penalizing model complexity, while advanced optimizers like AdamW and learning rate schedules improve convergence and generalization in neural network training.
Image Classification with Linear Classifiers
Linear classifiers use learned weight matrices and biases to assign class scores, enabling fast inference but only handling linearly separable data.
VLA-Fool
Researchers propose VLA-Fool, demonstrating how textual typos, visual patches, and cross-modal misalignment can adversarially attack vision-language-action models.
StableVLA
StableVLA introduces IB-Adapter, a plug-and-play module grounded in information bottleneck theory that enhances vision-language-action model robustness to visual corruptions without requiring extra training data.