Is it just a rebranding of the tnt abstraction? If not, how is it different, what new things can we expect? (Maybe im a bit impatient haha)
Thanks! 
1 Like
There is some more info on the ignite README now
Indeed it is very similar to tnt’s Engine (and was inspired by tnt). The main differences are:
Ignite allows for multiple handlers per event
Ignite doesn’t assume a single model to be trained (and has a higher level of abstraction), which makes it easier to train things like GANs
Ignite’s higher level of abstraction means that we require the user to define the closure to be run in the training and validation loop, whereas tnt creates it internally. This allows for a great deal more of flexibility, such as co-training multiple models and computing/tracking multiple losses and metrics.
Ignite is something we’ve been using for some time privately, and as tnt is not being actively developed for new features, we decided to push this to pytorch and develop it in the open.
That being said, there are some things from tnt we really like and would like to port over, such as the integration with Visdom (and possibly add integration with TensorBoard).
As always, PRs are welcome