MCPcopy Create free account
hub / github.com/Yang-Li-2000/Touch-Line-Transformer / update_ema

Function update_ema

util/optim.py:8–25  ·  view source on GitHub ↗

Apply exponential moving average update. The weights are updated in-place as follow: w_ema = w_ema * decay + (1 - decay) * w Args: model: active model that is being optimized model_ema: running average model decay: exponential decay parameter

(model, model_ema, decay)

Source from the content-addressed store, hash-verified

source not stored for this graph (policy: none)

Callers 1

train_one_epochFunction · 0.90

Calls

no outgoing calls

Tested by

no test coverage detected