diff --git a/python/sglang/eval/loogle_eval.py b/python/sglang/eval/loogle_eval.py
index 250f47a7b..22fc70541 100644
--- a/python/sglang/eval/loogle_eval.py
+++ b/python/sglang/eval/loogle_eval.py
@@ -142,7 +142,7 @@ if __name__ == "__main__":
     parser.add_argument(
         "--model",
         default="meta-llama/Llama-4-Maverick-17B-128E-Instruct",
-        help="Model name or ID",
+        help="Model name or ID, only used for model name",
     )
     parser.add_argument(
         "--max-concurrency", type=int, default=144, help="Maximum concurrent requests"
diff --git a/scripts/ci_install_dependency.sh b/scripts/ci_install_dependency.sh
index 9bd292a01..ae99fe4e0 100755
--- a/scripts/ci_install_dependency.sh
+++ b/scripts/ci_install_dependency.sh
@@ -31,3 +31,6 @@ pip install -e lmms-eval/
 
 # Install FlashMLA for attention backend tests
 pip install git+https://github.com/deepseek-ai/FlashMLA.git
+
+# Install hf_xet
+pip install huggingface_hub[hf_xet]
diff --git a/scripts/ci_install_dependency_8_gpu.sh b/scripts/ci_install_dependency_8_gpu.sh
index 6bd28c4c5..3167ff403 100755
--- a/scripts/ci_install_dependency_8_gpu.sh
+++ b/scripts/ci_install_dependency_8_gpu.sh
@@ -123,3 +123,6 @@ gdrcopy_copybw
 echo "=== Verify NVSHMEM ==="
 nvshmem-info -a
 # /opt/nvshmem/bin/perftest/device/pt-to-pt/shmem_put_bw
+
+# Install hf_xet
+pip install huggingface_hub[hf_xet]