ugh
This commit is contained in:
parent
cbf6b84e27
commit
f593ee98fc
|
@ -60,7 +60,7 @@ def load_engines(training=True, **model_kwargs):
|
||||||
|
|
||||||
# to handle the issue of training with deepspeed, but inferencing with local
|
# to handle the issue of training with deepspeed, but inferencing with local
|
||||||
if checkpoint_path.exists() and backend == "local":
|
if checkpoint_path.exists() and backend == "local":
|
||||||
tag = open(checkpoint_path).read()
|
tag = open(checkpoint_path).read().strip()
|
||||||
checkpoint_path = pick_path( checkpoint_path.parent / tag / f"state.{cfg.weights_format}", *[ f'.{format}' for format in cfg.supported_weights_formats] )
|
checkpoint_path = pick_path( checkpoint_path.parent / tag / f"state.{cfg.weights_format}", *[ f'.{format}' for format in cfg.supported_weights_formats] )
|
||||||
|
|
||||||
# if loaded using --model=
|
# if loaded using --model=
|
||||||
|
|
|
@ -258,7 +258,7 @@ class Engine():
|
||||||
self.optimizer.step()
|
self.optimizer.step()
|
||||||
self.optimizer.zero_grad()
|
self.optimizer.zero_grad()
|
||||||
|
|
||||||
self._get_grad_norm()
|
# self._get_grad_norm()
|
||||||
|
|
||||||
# doesn't actually work
|
# doesn't actually work
|
||||||
def _get_grad_norm(self):
|
def _get_grad_norm(self):
|
||||||
|
|
Loading…
Reference in New Issue
Block a user