5 papers
Tensor Memory: Fixed-Size Recurrent State for Long-Horizon Transformers
Kabir Swain, Sijie Han, Daniel Karl I. Weidele +2
Transformers process images and videos by flattening space and time into long token sequences. While attention and KV caching preserve past features, their memory grows with sequen…
Hurwitz Quaternion Multiplicative Quantization for KV Cache Compression
Kabir Swain, Sijie Han, Daniel Karl I. Weidele +3
We propose \textbf{Hurwitz Quaternion Multiplicative Quantization (HQMQ)}, a \textbf{calibration-free} method for KV cache compression of large language models. HQMQ treats each 4-…
Tensor Cache: Eviction-conditioned Associative Memory for Transformers
Kabir Swain, Sijie Han, Daniel Karl I. Weidele +2
Autoregressive Transformer KV caches grow linearly with context length; sliding-window caching bounds memory but discards evicted tokens entirely, so relevant evidence outside the…
ChartNet: A Million-Scale, High-Quality Multimodal Dataset for Robust Chart Understanding
Jovana Kondic, Pengyuan Li, Dhiraj Joshi +24
Understanding charts requires models to jointly reason over geometric visual patterns, structured numerical data, and natural language -- a capability where current vision-language…
Small Models, Smarter Learning: The Power of Joint Task Training
Csaba Both, Benjamin Hoover, Hendrik Strobelt +4
Multi-task learning improves generalization, but when does it reduce the model capacity required to learn? We provide a systematic study of how joint training affects the learning…