Enable bfloat16 training (coqui-ai#108)

* Enable bfloat16 training

* Bump up to v0.0.26

Bump up to v0.0.25

Bump up to v0.0.24

Save on exit (coqui-ai#102)

* Save on keyboard interrupt

* Bump up to v0.0.23

Bump up to v0.0.22

Custom optimize for handling complex trainings (coqui-ai#89)

* Implement `isimplemened`

* Replace `hasattr` with `isimplemented`

* Add `torch.set_grad_emabled` toggles

* Implement custom optimize

* Update model

* Add GAN training tests

* Make style

* Make lint

* Do not expect `grad_norm` returned

* Add training examples

* Bump up to v0.0.21

Bump up to v0.0.20

Bump up to v0.0.19

Add start with eval option (coqui-ai#84)

* Add start with eval option

* Ping for training run

* Drop p3.6 from CI

* Turn off telemetry on CI

* Bump up to v0.0.18

Bump up to v0.0.17