GRADMM
A gradient-based synthetic data generation method for LLM fine-tuning that generates training data by reconstructing source text or pursuing linguistic fluency. It serves as a comparison baseline for the DASA method.
Timeline (last 90 days)
DASA outperforms the baseline method GRADMM in most comparisons.
Under the synthetic settings evaluated, DASA provides a 3.6 to 4.9× speedup compared to GRADMM, with comparable peak GPU memory usage.
The comparison baseline GRADMM tends to attempt to reconstruct the source text or pursue linguistic fluency.
All Facts (3)
DASA outperforms the baseline method GRADMM in most comparisons.
50%UnverifiedUnder the synthetic settings evaluated, DASA provides a 3.6 to 4.9× speedup compared to GRADMM, with comparable peak GPU memory usage.
50%UnverifiedThe comparison baseline GRADMM tends to attempt to reconstruct the source text or pursue linguistic fluency.
50%