zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-06-26 10:41:18 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Cleaner masking code

Browse Source
This commit is contained in:
Joe Barrow 2023-12-09 21:21:24 -05:00
parent 5d4838b02e
commit a577abc313
1 changed files with 1 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
bert/model.py
View File

@ -123,22 +123,13 @@ class Bert(nn.Module):
if attention_mask is not None: if attention_mask is not None:
# convert 0's to -infs, 1's to 0's, and make it broadcastable # convert 0's to -infs, 1's to 0's, and make it broadcastable
attention_mask = self.convert_mask_to_additive_causal_mask(attention_mask) attention_mask = mx.log(attention_mask)
attention_mask = mx.expand_dims(attention_mask, (1, 2)) attention_mask = mx.expand_dims(attention_mask, (1, 2))
y = self.encoder(x, attention_mask) y = self.encoder(x, attention_mask)
return y, mx.tanh(self.pooler(y[:, 0])) return y, mx.tanh(self.pooler(y[:, 0]))
def convert_mask_to_additive_causal_mask(
self, mask: mx.array, dtype: mx.Dtype = mx.float32
) -> mx.array:
mask = mask == 0
mask = mask.astype(dtype) * -1e9
return mask
def load_model(bert_model: str, weights_path: str) -> tuple[Bert, BertTokenizer]: def load_model(bert_model: str, weights_path: str) -> tuple[Bert, BertTokenizer]:
# load the weights npz # load the weights npz
weights = mx.load(weights_path) weights = mx.load(weights_path)