Commit 1e1bd6b
authored
Make fp16 flash attention fallback to mkl implementation on platforms that do not support amx fp16 (#3386)
* make fp16 flash attention fallback to mkl implementation on platforms that do not support amx fp16
* fix format
* add amx check for bf16
* upgrade mkl version and minor changes
* minor changes
* minor fix
* add vnni support for bf16 odd shapes1 parent f1aa2de commit 1e1bd6b
File tree
4 files changed
+307
-245
lines changed- cmake/Modules
- csrc/cpu
- aten
- kernels
- utils
- tpp
4 files changed
+307
-245
lines changed| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
35 | 35 | | |
36 | 36 | | |
37 | 37 | | |
38 | | - | |
| 38 | + | |
39 | 39 | | |
40 | 40 | | |
41 | 41 | | |
| |||
0 commit comments