Jiaxin Gu

Why not do log_softmax("arch_param") in the graph？

4

In [train_search.py](https://github.com/AberHu/TF-NAS/blob/35a34a11b6a64ecf1047cb7acb016e04f99ea259/train_search.py#L422), I noticed that you do log_softmax() out of the graph, but why? Why not just use param **alpha** instead and do log_softmax() in each forward step?

Nice work! But could you provide the pertained models and downstream models for evaluation?

Will you guys release the Objects365 pretraining models for InternImage-XL/H DINO in the Table 4?