diff --git a/integration-tests/conftest.py b/integration-tests/conftest.py
index 69b3fe67..efeda08d 100644
--- a/integration-tests/conftest.py
+++ b/integration-tests/conftest.py
@@ -319,7 +319,7 @@ def launcher(event_loop):
 
         env = {
             "LOG_LEVEL": "info,text_generation_router=debug",
-            "ENABLE_CUDA_GRAPHS": "True",
+            "ENABLE_CUDA_GRAPHS": "true",
         }
         if not use_flash_attention:
             env["USE_FLASH_ATTENTION"] = "false"
diff --git a/server/text_generation_server/models/flash_causal_lm.py b/server/text_generation_server/models/flash_causal_lm.py
index 21ed4f6c..e68a2100 100644
--- a/server/text_generation_server/models/flash_causal_lm.py
+++ b/server/text_generation_server/models/flash_causal_lm.py
@@ -777,7 +777,7 @@ class FlashCausalLM(Model):
             self.device,
         )
 
-        if os.getenv("ENABLE_CUDA_GRAPHS", "False") == "True":
+        if os.getenv("ENABLE_CUDA_GRAPHS", "false") == "true":
             try:
                 # Warmup cuda graphs for all power of twos until 64
                 for i in range(6):