Skip to content

[None][fix] Avoid unnecessary concat in attn_output_gate case. #8094

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
+4 −4
Open

[None][fix] Avoid unnecessary concat in attn_output_gate case. #8094

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 2 additions & 3 deletions tensorrt_llm/_torch/modules/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -537,10 +537,9 @@ def forward(
t.reshape(*orig_shape, -1) for t in torch.chunk(
q_gate.view(*orig_shape, self.num_heads, -1), 2, dim=-1)
]
### TODO: avoid the redundant split and concat
qkv = torch.concat([q, k, v], dim=-1)
else:
q, k, v = qkv, None, None

q, k, v = qkv, None, None
q, k, v = self.apply_rope(q, k, v, position_ids)
q, k, v = self.convert_qkv(q, k, v)

Expand Down
3 changes: 2 additions & 1 deletion tensorrt_llm/_torch/modules/qk_norm_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -249,6 +249,7 @@ def apply_rope(self, q: torch.Tensor, k: Optional[torch.Tensor],
else:
return q, k, v

assert k is None and v is None, "The input should be a concatenated qkv tensor to apply_qk_norm_rope"
qkv = q
if k is not None and v is not None:
qkv = torch.concat([q, k, v], dim=-1)
return self.apply_qk_norm_rope(qkv, position_ids)