tfixup_init_#
- core.nn.tfixup_init_(encoder, scale_base, v_scale_factor)[source]#
Scale a Transformer encoder’s weights in place for optimization without warmup.
Sources: Transformers without Tears https://arxiv.org/pdf/1910.05895.pdf T-Fixup http://www.cs.toronto.edu/~mvolkovs/ICML2020_tfixup.pdf
- Parameters:
encoder (
TransformerEncoder) – encoder whoselayersuse the stocknn.TransformerEncoderLayerparameter names (linear1/linear2/self_attn).scale_base (
float) – T-Fixup base scale, applied asscale_base * N**-0.25.v_scale_factor (
float) – extra factor on the value weights, on top ofscale.
- Return type: