[Docs] Refactor the attention backend docs/skill (#3475)
Signed-off-by: gcanlin <canlinguosdu@gmail.com> Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
This commit is contained in:
1 parent
67e0e10c85
commit
ef7f2f1cd2
6 files changed
+186
-66
No files matched your search
@@ -92,6 +92,7 @@ class TPSelfAttention(nn.Module):
|
||||
softmax_scale=1.0 / (self.head_dim ** 0.5),
|
||||
causal=False,
|
||||
num_kv_heads=self.to_qkv.num_kv_heads, # Local KV heads per GPU
|
||||
role="self",
|
||||
)
|
||||
|
||||
def forward(self, x):
|
||||
|
||||
Reference in new issue
Block a user