Darwin Family: VIDRAFT's Training-Free Model Merging Framework Hits 86.9% on GPQA Diamond
Darwin Family: VIDRAFT's Training-Free Model Merging Framework Hits 86.9% on GPQA Diamond TL;DR: VIDRAFT, a Korean AI startup, has published a framework called Darwin Family that improves LLM reasoning ability through evolutionary-algorithm-guided parameter recombination — no gradient training required. Their flagship model, Darwin-27B-Opus, scored 86.9% on GPQA Diamond and ranked 6th globally…
Darwin Family, an innovative training-free model merging framework developed by the VIDRAFT research team in Seoul, Korea, has achieved remarkable results with a 86.9% score on the GPQA Diamond benchmark. This groundbreaking technology combines the weight parameters of two existing models, one generalist and one specialized in reasoning, to create a new model that inherits the best capabilities of both parents, all without any additional gradient training.
The framework, named after Charles Darwin, employs a unique evolutionary search process to optimize the merging of model parameters based on a sophisticated MRI Diagnostic Importance Scoring system.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
