LLaDA-8B-Instruct
LLaDA-8B-Instruct is an instruction-tuned version of an 8-billion-parameter diffusion language model (dLLM) that can be used for text generation and reasoning tasks.
Timeline (last 90 days)
The experiments were conducted on two models, LLaDA-8B-Instruct and Dream-v0-Instruct-7B, covering five benchmarks.
Compared with direct continuous-time sampling, SpecCTMC reduces the number of sequential model evaluations by 1.73 to 4.38 times while maintaining the reference sampling distribution.
On GSM8K, dSpark achieves a 10.12x speedup on LLaDA-8B-Instruct relative to multiple baseline methods.
All Facts (3)
The experiments were conducted on two models, LLaDA-8B-Instruct and Dream-v0-Instruct-7B, covering five benchmarks.
50%UnverifiedCompared with direct continuous-time sampling, SpecCTMC reduces the number of sequential model evaluations by 1.73 to 4.38 times while maintaining the reference sampling distribution.
50%UnverifiedOn GSM8K, dSpark achieves a 10.12x speedup on LLaDA-8B-Instruct relative to multiple baseline methods.
50%