CadLLM accelerates diffusion-based LLMs up to 2.28x in throughput via confidence-aware adaptive control of generation parameters and sampling breadth while maintaining competitive accuracy.
2) Eggs eaten for breakfast and baked for friends:3 + 4 = 7
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibration
CadLLM accelerates diffusion-based LLMs up to 2.28x in throughput via confidence-aware adaptive control of generation parameters and sampling breadth while maintaining competitive accuracy.