ICLR conference banner - abstract shapes
2024年5月7日 – 2024年5月11日

Microsoft at ICLR 2024

Central European Time (UTC +1)

地点： Vienna, Austria

微软研究院博客

White ICLR logo to the left of the first page of the accepted paper, “Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs” on a purple background.
微软研究院博客

| Liyuan Liu 和 Jianfeng Gao

LLMs rely on memory-intensive mechanisms like the key-value (KV) cache to store and quickly retrieve data. FastGen optimizes KV cache usage, reducing LLM memory demands by up to 50% while maintaining performance.

LoftQ paper at ICLR 2024
微软研究院博客

| Nikos Karampatziakis, Chen Liang, Weizhu Chen, Yixiao Li, Yifan Yu, 和 Tuo Zhao

LoftQ boosts LLM efficiency by streamlining the fine-tuning process, reducing computational demands while preserving high performance. Innovations like this can help make AI technology more energy-efficient.

Stylized microphone and sound waves illustration.
微软研究院播客

| Michel Galley 和 Gretchen Huizinga

Researcher Michel Galley explores how he and fellow researchers combined new and existing data to create MathVista, an open-source benchmark for measuring the mathematical reasoning capabilities of foundation models in scenarios that involve text and images.