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

Update mistral.py

Browse source
This commit is contained in:
Daniel Han-Chen 2024-01-21 19:01:28 +11:00
commit 5c927e4930
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
unsloth/models/mistral.py
View file

@ -47,10 +47,9 @@ def MistralAttention_fast_forward(
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
bsz, q_len, _ = hidden_states.size()
Q, K, V = self.apply_qkv(self, hidden_states)
# Check for inference
if use_cache and past_key_value is not None and q_len == 1:
if past_key_value is not None and q_len == 1:
A, past_key_value = LlamaAttention_fast_forward_inference(
self,
hidden_states,
@ -66,6 +65,7 @@ def MistralAttention_fast_forward(
head_dim = self.head_dim
assert(n_kv_heads * n_groups == n_heads)
Q, K, V = self.apply_qkv(self, hidden_states)
Q = Q.view(bsz, q_len, n_heads, head_dim).transpose(1, 2)
K = K.view(bsz, q_len, n_kv_heads, head_dim).transpose(1, 2)
V = V.view(bsz, q_len, n_kv_heads, head_dim).transpose(1, 2)