Search papers, labs, and topics across Lattice.
Central South University
3
0
7
Even the best large vision-language models struggle with multi-image reasoning, scoring only 50% on a new benchmark designed to challenge their capabilities.
LLMs can be aggressively quantized to W(1+1)A4 without significant performance degradation using a surprisingly simple three-stage distillation approach.
Merging LoRA adapters can actually enhance performance when treating their matrices separately, with Pico achieving up to 8.3 points higher accuracy than existing methods.
