Update README.md

import torch
from torch import nn
# Definir la red, lo que en definitiva la hace siamesa es pasar los dos dic
class SBERT(nn.Module):
def __init__(self, base_model, dropout=0.1):
super().__init__()
self.base_model = base_model
self.dropout = nn.Dropout(dropout)
# Recordamos que la salida de la Bert es 768
#self.fc = nn.Linear(768, 3) #Cambio 13/6
self.fc = nn.Linear(768*3, 3)

def forward(self, premise, hypothesis):
out_u = self.base_model(**premise)
out_v = self.base_model(**hypothesis)
pooler_u = out_u.pooler_output
pooler_v = out_v.pooler_output
pooler_u = self.dropout(pooler_u)
pooler_v = self.dropout(pooler_v)
#concatenated = torch.cat([self.fc(pooler_u), self.fc(pooler_v), torch.abs(self.fc(pooler_u) - self.fc(pooler_v))], dim=0)
concatenated = torch.cat([pooler_u, pooler_v, torch.abs(pooler_u -pooler_v)], dim=1)
out=self.fc(concatenated)
return out

README.md

@@ -8,26 +8,3 @@ pipeline_tag: fill-mask
 tags:
 - code
 ---
-import torch
-from torch import nn
-# Definir la red,  lo que en definitiva la hace siamesa es pasar los dos dic
-class SBERT(nn.Module):
-    def __init__(self, base_model, dropout=0.1):
-        super().__init__()
-        self.base_model = base_model
-        self.dropout = nn.Dropout(dropout)
-        # Recordamos que la salida de la Bert es 768
-        #self.fc = nn.Linear(768, 3) #Cambio 13/6
-        self.fc = nn.Linear(768*3, 3)
-
-    def forward(self, premise, hypothesis):
-        out_u = self.base_model(**premise)
-        out_v = self.base_model(**hypothesis)
-        pooler_u = out_u.pooler_output
-        pooler_v = out_v.pooler_output
-        pooler_u = self.dropout(pooler_u)
-        pooler_v = self.dropout(pooler_v)
-        #concatenated = torch.cat([self.fc(pooler_u), self.fc(pooler_v), torch.abs(self.fc(pooler_u) - self.fc(pooler_v))], dim=0)
-        concatenated = torch.cat([pooler_u, pooler_v, torch.abs(pooler_u -pooler_v)], dim=1)
-        out=self.fc(concatenated)
-        return out