Add support for loading torchao models with AOPerModuleConfig (#17826)

Signed-off-by: Jerry Zhang <jerryzh168@gmail.com>

tests/quantization/test_torchao.py

@@ -31,9 +31,6 @@ def test_pre_quantized_model(vllm_runner):
     ])
 def test_opt_125m_int4wo_model_loading_with_params(vllm_runner,
                                                    pt_load_map_location):
-    """
-    Test loading roberta-base model with no lm_head.
-    """
     torch._dynamo.reset()
     model_name = "jerryzh168/opt-125m-int4wo"
     with vllm_runner(model_name=model_name,
@@ -47,5 +44,20 @@ def test_opt_125m_int4wo_model_loading_with_params(vllm_runner,
         print(output)
 
 
+@pytest.mark.skipif(not TORCHAO_AVAILABLE, reason="torchao is not available")
+def test_opt_125m_int4wo_model_per_module_quant(vllm_runner):
+    torch._dynamo.reset()
+    model_name = "jerryzh168/opt-125m-int4wo-per-module"
+    with vllm_runner(model_name=model_name,
+                     quantization="torchao",
+                     dtype="bfloat16",
+                     pt_load_map_location="cuda:0") as llm:
+        output = llm.generate_greedy(["The capital of France is"],
+                                     max_tokens=32)
+
+        assert output
+        print(output)
+
+
 if __name__ == "__main__":
     pytest.main([__file__])