issues
search
world-modelz
/
dreamax
A scalable Dreamer implementation in JAX
MIT License
11
stars
2
forks
source link
Improve serialization of model & resumption of training
#11
Open
andreaskoepf
opened
2 years ago
andreaskoepf
commented
2 years ago
it should be possible to resume model training (all relevant model and optimizer parameters must be serialized)
different model parts could be serialized individually
model checkpoints should be written in configurable intervals (e.g. every N gradient steps)