1 paper
Menghao Li, Linjie Mu, Yin Wang +6
Autoregressive vision language models unify heterogeneous perception tasks but are highly susceptible to compounding errors. On-policy distillation (OPD) bridges the training-infer…