Search papers, labs, and topics across Lattice.
Shaanxi Normal University
3
0
7
MLLMs struggle to convert visual evidence into context-specific actions, with performance plummeting by over 44% in certain scenarios.
AutoPass achieves up to 1.117x speedup in compiler performance tuning by allowing LLMs to interact with compiler internals, challenging the notion that black-box approaches are sufficient.
Static GPU partitioning alone can't solve underutilization, but fine-grained CPU offloading over Nvlink-C2C can bridge the gap.
