Flash-dLLM: IO-Aware KV Caching and Parallel Decoding for Fast, Memory-Efficient Diffusion LLMs
Open paper
Benchmark
Model
Rank
Results
code-generation-on-humaneval
Flash-dLLM
#111
Pass@1: 40.24
math-word-problem-solving-on-math
Flash-dLLM
#125
Accuracy: 36.56
Rank counts only results with a code link.