Tag: model compression
Browse our exclusive articles!
Spectral Compact Training for Efficient Large Language Models
Discover Spectral Compact Training, a memory-efficient method enabling large language model training on consumer hardware with truncated SVD and Stiefel QR...
TED: Training-Free Knowledge Distillation for Multimodal AI
Discover TED, a training-free knowledge distillation method that boosts multimodal reasoning performance without costly model updates or large datasets.
Edge Reliability Challenges in Compressed Vision-Language Models
Explore failure modes of compressed vision-language models under visual corruption and their impact on edge deployment reliability.
DUET-VLM: Efficient Dual-Stage Token Reduction for VLMs
Discover DUET-VLM, a dual-stage token reduction method boosting Vision-Language Model efficiency with minimal accuracy loss.
Efficient Hybrid Sequence Models via Generation-Focused Distillation
Discover how generation-focused distillation improves hybrid sequence models, reducing inference costs while maintaining high-quality output generation.