Support tensor parallel (#2)
This commit is contained in:
@@ -112,7 +112,7 @@ class OPTCacheFlowAttention(nn.Module):
|
||||
output[:num_prompt_tokens],
|
||||
query[:num_prompt_tokens],
|
||||
key[:num_prompt_tokens],
|
||||
value[:num_prompt_tokens],
|
||||
value[:num_prompt_tokens],
|
||||
input_metadata.prompt_lens,
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user