def randw(*d):
return nn.Parameter(torch.empty(*d).normal_(0, 1 / math.sqrt(d[-1])))
- self.wq = randw(nb_heads, dim_qk, dim_in)
- self.wk = randw(nb_heads, dim_qk, dim_in)
- self.wv = randw(nb_heads, dim_v, dim_in)
+ self.w_q = randw(nb_heads, dim_qk, dim_in)
+ self.w_k = randw(nb_heads, dim_qk, dim_in)
+ self.w_v = randw(nb_heads, dim_v, dim_in)
self.causal = causal
self.attention_dropout = attention_dropout
def forward(self, x):
- q = torch.einsum('ntc,hdc->nhtd', x, self.wq)
- k = torch.einsum('ntc,hdc->nhtd', x, self.wk)
- v = torch.einsum('ntc,hdc->nhtd', x, self.wv)
+ q = torch.einsum('ntc,hdc->nhtd', x, self.w_q)
+ k = torch.einsum('ntc,hdc->nhtd', x, self.w_k)
+ v = torch.einsum('ntc,hdc->nhtd', x, self.w_v)
r = math.sqrt(q.size(3))
a = torch.einsum('nhtd,nhsd->nhts', q, k).div(r)
if self.causal:
return x
######################################################################
+
+if __name__ == '__main__':
+ vocabulary_size = 10
+ x = torch.randint(vocabulary_size, (25, 100))
+
+ model = MyGPT(
+ vocabulary_size = vocabulary_size,
+ dim_model = 16, dim_keys = 50, dim_hidden = 100,
+ nb_heads = 2, nb_blocks = 3,
+ dropout = 0.1
+ )
+
+ y = model(x)
+
+######################################################################