Kevin P Murphy
Results
112
issues of
Kevin P Murphy
I don't see Student T listed at https://www.gen.dev/docs/dev/ref/modeling/distributions/#Built-In-Distributions. It is very useful for making robust models. (You have Cauchy, which is a special case with dof=1)
The default settings in https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/recipes/multiplayer_rl/text_arena/train.py use `num_train_datapoints = 131072` with a batch size of 512, which amounts to 256 steps. However, I found that you can get good results with...