fix: add back typo removed variable

huggingface · Feb 21, 2024 · 66f8912 · 66f8912
1 parent 3e22cdd
commit 66f8912
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/server/text_generation_server/models/__init__.py b/server/text_generation_server/models/__init__.py
@@ -43,6 +43,8 @@
 
 FLASH_ATT_ERROR_MESSAGE = "{} requires Flash Attention enabled models."
 
+FLASH_ATTENTION = True
+
 # FlashCausalLM reqiures CUDA Graphs to be enabled on the system. This will throw a RuntimeError
 # if CUDA Graphs are not available when calling `torch.cuda.graph_pool_handle()` in the FlashCausalLM
 HAS_CUDA_GRAPH = False