Try to make cuda call non-blocking

This commit is contained in:
Matthw Honnibal 2019-10-18 17:22:16 +02:00
parent b2e8f37965
commit 4da1c1c211

View File

@ -263,7 +263,7 @@ class ParserStepModel(Model):
self.state2vec = precompute_hiddens(len(docs), self.tokvecs, layers[1], self.state2vec = precompute_hiddens(len(docs), self.tokvecs, layers[1],
drop=drop) drop=drop)
self.vec2scores = layers[-1] self.vec2scores = layers[-1]
self.cuda_stream = util.get_cuda_stream() self.cuda_stream = util.get_cuda_stream(non_blocking=True)
self.backprops = [] self.backprops = []
self._class_mask = numpy.zeros((self.vec2scores.nO,), dtype='f') self._class_mask = numpy.zeros((self.vec2scores.nO,), dtype='f')
self._class_mask.fill(1) self._class_mask.fill(1)