move vllm/flash-attn to separate dep group

Author: gautierdag

pyproject.toml

@@ -12,6 +12,7 @@ dependencies = [
     "loguru",
     "networkx",
     "openai",
+    "litellm>=1.71.1",
     "pandas",
     "pillow>=8.0.0",
     "pydantic",
@@ -46,10 +47,11 @@ full = [
     "torchvision>=0.20.0",
     "einops",
     "huggingface-hub",
-    "transformers==4.52.3",
+    "transformers==4.52.3",   
+]
+serve = [
     "flash-attn>=2.7.4.post1",
     "flashinfer-python==0.2.2; sys_platform != 'darwin'",
-    "litellm>=1.71.1",
     "vllm>=0.8.5",
 ]