Commit
·
8191fb5
1
Parent(s):
47450e4
Upload LUAR
Browse files
model.py
CHANGED
@@ -16,11 +16,15 @@ class SelfAttention(nn.Module):
|
|
16 |
super(SelfAttention, self).__init__()
|
17 |
|
18 |
def forward(self, k, q, v):
|
19 |
-
|
20 |
-
|
21 |
-
|
|
|
|
|
|
|
|
|
22 |
|
23 |
-
|
24 |
|
25 |
class LUAR(PreTrainedModel):
|
26 |
"""Defines the LUAR model.
|
@@ -85,4 +89,4 @@ class LUAR(PreTrainedModel):
|
|
85 |
"""
|
86 |
output = self.get_episode_embeddings(input_ids, attention_mask, output_attentions)
|
87 |
|
88 |
-
return output
|
|
|
16 |
super(SelfAttention, self).__init__()
|
17 |
|
18 |
def forward(self, k, q, v):
|
19 |
+
if hasattr(F, "scaled_dot_product_attention") and torch.cuda.is_available():
|
20 |
+
with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_mem_efficient=True):
|
21 |
+
return F.scaled_dot_product_attention(k, q, v)
|
22 |
+
else:
|
23 |
+
d_k = q.size(-1)
|
24 |
+
scores = torch.matmul(k, q.transpose(-2, -1)) / math.sqrt(d_k)
|
25 |
+
p_attn = F.softmax(scores, dim=-1)
|
26 |
|
27 |
+
return torch.matmul(p_attn, v)
|
28 |
|
29 |
class LUAR(PreTrainedModel):
|
30 |
"""Defines the LUAR model.
|
|
|
89 |
"""
|
90 |
output = self.get_episode_embeddings(input_ids, attention_mask, output_attentions)
|
91 |
|
92 |
+
return output
|