diff --git a/docs/source/models/supported_models.rst b/docs/source/models/supported_models.rst
index 11c122f839b10..3dcc242803752 100644
--- a/docs/source/models/supported_models.rst
+++ b/docs/source/models/supported_models.rst
@@ -109,7 +109,7 @@ Decoder-only Language Models
     -
   * - :code:`MiniCPM3ForCausalLM`
     - MiniCPM3
-    - :code:`openbmb/MiniCPM3-4B`
+    - :code:`openbmb/MiniCPM3-4B`, etc.
     -
   * - :code:`MistralForCausalLM`
     - Mistral, Mistral-Instruct
diff --git a/tests/models/decoder_only/language/test_big_models.py b/tests/models/decoder_only/language/test_big_models.py
index c5c10eeecdea1..5168210f15013 100644
--- a/tests/models/decoder_only/language/test_big_models.py
+++ b/tests/models/decoder_only/language/test_big_models.py
@@ -5,7 +5,8 @@
 Run `pytest tests/models/test_big_models.py`.
 """
 import pytest
-import torch
+
+from vllm.platforms import current_platform
 
 from ...utils import check_outputs_equal
 
@@ -21,9 +22,7 @@
 ]
 
 #TODO: remove this after CPU float16 support ready
-target_dtype = "float"
-if torch.cuda.is_available():
-    target_dtype = "half"
+target_dtype = "float" if current_platform.is_cpu() else "half"
 
 
 @pytest.mark.parametrize("model", MODELS)
@@ -37,6 +36,9 @@ def test_models(
     dtype: str,
     max_tokens: int,
 ) -> None:
+    if model.startswith("openbmb/MiniCPM3") and current_platform.is_cpu():
+        pytest.skip("MiniCPM requires fused_moe which is not supported by CPU")
+
     with hf_runner(model, dtype=dtype) as hf_model:
         hf_outputs = hf_model.generate_greedy(example_prompts, max_tokens)