mirror of https://github.com/explosion/spaCy.git
b2ef6100af
Previously, pipelines with shared tok2vec weights would call the tok2vec backprop callback multiple times, once for each pipeline component. This caused errors for PyTorch, and was inefficient. Instead, accumulate the gradient for all but one component, and just call the callback once. |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| entityruler.py | ||
| functions.py | ||
| hooks.py | ||
| morphologizer.pyx | ||
| pipes.pyx | ||
| tok2vec.py | ||