memory footprint
The amount of memory that an AI model or process consumes during execution, which is critical for deployment in resource-constrained environments.
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers
- GeoLLaVA-8K: Scaling Remote-Sensing Multimodal Large Language Models to 8K Resolution
- Hardware-aligned Hierarchical Sparse Attention for Efficient Long-term Memory Access
- Multi-head Temporal Latent Attention
- Q-Palette: Fractional-Bit Quantizers Toward Optimal Bit Allocation for Efficient LLM Deployment
- Small Batch Size Training for Language Models: When Vanilla SGD Works, and Why Gradient Accumulation is Wasteful
- Taxonomy of reduction matrices for Graph Coarsening
- Test-Time Spectrum-Aware Latent Steering for Zero-Shot Generalization in Vision-Language Models