Use flash attention monkeypatch
This commit is contained in:
@@ -16,6 +16,9 @@
|
|||||||
}
|
}
|
||||||
]
|
]
|
||||||
"""
|
"""
|
||||||
|
from monkeypatch.llama_flash_attn_monkey_patch import replace_llama_attn_with_flash_attn
|
||||||
|
|
||||||
|
replace_llama_attn_with_flash_attn()
|
||||||
|
|
||||||
import sys
|
import sys
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user