support torchair mode (#2641)

### What this PR does / why we need it?
support torchair mode
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?

- vLLM version: v0.10.1.1
- vLLM main:
5438967fbc

Signed-off-by: zhangdepeng <zhangdepeng2@huawei.com>
Signed-off-by: p00465316 <panchao13@huawei.com>
Co-authored-by: zhangdepeng <zhangdepeng2@huawei.com>

tests/ut/test_ascend_config.py

@@ -46,6 +46,7 @@ class TestAscendConfig(TestBase):
 
         torchair_graph_config = ascend_config.torchair_graph_config
         self.assertFalse(torchair_graph_config.enabled)
+        self.assertEqual(torchair_graph_config.mode, '')
         self.assertFalse(torchair_graph_config.use_cached_graph)
         self.assertEqual(torchair_graph_config.graph_batch_sizes, [])
         self.assertFalse(torchair_graph_config.graph_batch_sizes_init)
@@ -294,6 +295,17 @@ class TestAscendConfig(TestBase):
             }
             init_ascend_config(test_vllm_config)
 
+        # mode should not be configured without torchair graph mode
+        with self.assertRaises(RuntimeError):
+            test_vllm_config.additional_config = {
+                "torchair_graph_config": {
+                    "enabled": False,
+                    "mode": 'max-autotune',
+                },
+                "refresh": True
+            }
+            init_ascend_config(test_vllm_config)
+
         # enable_kv_nz should not be enabled without torchair graph mode
         with self.assertRaises(RuntimeError):
             test_vllm_config.additional_config = {