Hello,
I am trying to reproduce the numbers from Table 1 in the paper, but the only ImageNet256 models that are available on hugging face are tiny, small and base, with base being the largest one. However, when the base model is loaded, and a summary of its parameters is printed out, it says that it has 142.225M parameters, which is significantly less than the sizes reported in Table 1, i.e.:
- 484M (which seems to correspond to the
large version of the model, only available for resolution 384 on hugging face), and
- 705M.
Can the authors share the models corresponding to the ones used to report results in Table 1 in the paper?
Hello,
I am trying to reproduce the numbers from Table 1 in the paper, but the only ImageNet256 models that are available on hugging face are
tiny,smallandbase, withbasebeing the largest one. However, when thebasemodel is loaded, and a summary of its parameters is printed out, it says that it has 142.225M parameters, which is significantly less than the sizes reported in Table 1, i.e.:largeversion of the model, only available for resolution 384 on hugging face), andCan the authors share the models corresponding to the ones used to report results in Table 1 in the paper?