fix: typos (#18151)
Signed-off-by: omahs <73983677+omahs@users.noreply.github.com>
This commit is contained in:
@@ -172,7 +172,7 @@ __device__ void paged_attention_kernel(
|
||||
|
||||
// Load the query to registers.
|
||||
// Each thread in a thread group has a different part of the query.
|
||||
// For example, if the the thread group size is 4, then the first thread in
|
||||
// For example, if the thread group size is 4, then the first thread in
|
||||
// the group has 0, 4, 8, ... th vectors of the query, and the second thread
|
||||
// has 1, 5, 9, ... th vectors of the query, and so on. NOTE(woosuk): Because
|
||||
// q is split from a qkv tensor, it may not be contiguous.
|
||||
@@ -259,7 +259,7 @@ __device__ void paged_attention_kernel(
|
||||
|
||||
// Load a key to registers.
|
||||
// Each thread in a thread group has a different part of the key.
|
||||
// For example, if the the thread group size is 4, then the first thread in
|
||||
// For example, if the thread group size is 4, then the first thread in
|
||||
// the group has 0, 4, 8, ... th vectors of the key, and the second thread
|
||||
// has 1, 5, 9, ... th vectors of the key, and so on.
|
||||
for (int i = 0; i < NUM_TOKENS_PER_THREAD_GROUP; i++) {
|
||||
|
||||
Reference in New Issue
Block a user