ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update llama.py

Browse source
This commit is contained in:
Daniel Han-Chen 2024-02-07 19:25:46 +11:00
commit e94647c2dc
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
unsloth/models/llama.py
View file

@ -199,7 +199,7 @@ def LlamaAttention_fast_forward_inference(
A = torch.matmul(Qn, Knn.transpose(2, 3)) #out = self.attention[:,:,:,:attention_size])
A *= self.scalar
A[:] = torch.nn.functional.softmax(A, dim = -1, dtype = torch.float32)#.to(A.dtype)
A = torch.matmul(A, Vnn, out = Qn)
A = torch.matmul(A, Vnn)#, out = Qn)
A = A.transpose(1, 2)
A = A.reshape(bsz, 1, self.hidden_size)
A = fast_linear_forward(self.o_proj, A, out = self.temp_QA[1])