diff --git a/unsloth/kernels/flex_attention.py b/unsloth/kernels/flex_attention.py index 7deb9a96d3..887ffca1b7 100644 --- a/unsloth/kernels/flex_attention.py +++ b/unsloth/kernels/flex_attention.py @@ -15,6 +15,7 @@ import torch from functools import lru_cache from transformers.models.llama.modeling_llama import logger +import os torch_compile_options = { "epilogue_fusion" : True,