MCPcopy Create free account
hub / github.com/CompVis/zigma / update_ema

Function update_ema

utils/train_utils.py:104–113  ·  view source on GitHub ↗

Step the EMA model towards the current model.

(ema_model, model, decay=0.9999)

Source from the content-addressed store, hash-verified

102
103@torch.no_grad()
104def update_ema(ema_model, model, decay=0.9999):
105 """
106 Step the EMA model towards the current model.
107 """
108 ema_params = OrderedDict(ema_model.named_parameters())
109 model_params = OrderedDict(model.named_parameters())
110
111 for name, param in model_params.items():
112 # TODO: Consider applying only to params that require_grad to avoid small numerical changes of pos_embed
113 ema_params[name].mul_(decay).add_(param.data, alpha=1 - decay)
114
115
116def requires_grad(model, flag=True):

Callers 1

mainFunction · 0.90

Calls

no outgoing calls

Tested by

no test coverage detected