Hello! I have a modified FFCV branch which I have been using to load a model from a checkpoint (to avoid mid-training disconnect/random failures), as well as validate the final_weights.pt loaded from disk (eg, if we wish to run it on another val.ffcv file).
I would gladly raise a PR for review for this, if you'd like this feature to be added.
Hello! I have a modified FFCV branch which I have been using to load a model from a checkpoint (to avoid mid-training disconnect/random failures), as well as validate the final_weights.pt loaded from disk (eg, if we wish to run it on another val.ffcv file).
I would gladly raise a PR for review for this, if you'd like this feature to be added.