Stanford University: AI Research, FlashAttention and DPO

An institution across several generations of AI

Stanford was founded in 1885 and opened to students in 1891. Its AI laboratory dates to 1963, with John McCarthy and Les Earnest named in the university's history. The later HAI institute connects technical work with education and policy across Stanford's schools. [1][4]

FlashAttention studies the memory traffic required to compute attention. Its tiled algorithm computes exact attention while reducing transfers between GPU memory and faster on-chip storage; the contribution is not an approximation of the attention operation. [2]

Direct Preference Optimization recasts preference-based language-model training as a direct objective, avoiding a separately fitted reward model and reinforcement-learning optimization loop in that method. These research artifacts help explain Stanford's role in model training, alongside the institutional work of HAI. [3][4]