support torchair mode (vllm-project#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:
vllm-project/vllm@5438967

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

Author: panchao-hub

docs/source/user_guide/configuration/additional_config.md

@@ -43,6 +43,7 @@ The details of each config option are as follows:
 | Name | Type | Default | Description |
 | ---- | ---- | ------- | ----------- |
 | `enabled` | bool | `False` | Whether to enable torchair graph mode. Currently only DeepSeek series models and PanguProMoE are supported to use torchair graph mode |
+| `mode` | str | `None` | When using reduce-overhead mode for torchair, mode needs to be set |
 | `enable_multistream_mla`| bool | `False` | Whether to put vector ops of MLA to another stream. This option only takes effects on models using MLA (e.g., DeepSeek). |
 | `enable_multistream_moe`| bool | `False` | Whether to enable multistream shared expert. This option only takes effects on DeepSeek moe models. |
 | `enable_view_optimize` | bool | `True` | Whether to enable torchair view optimization |

tests/ut/test_ascend_config.py

@@ -46,6 +46,7 @@ def test_init_ascend_config_without_additional_config(self):
 
         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 @@ def test_ascend_config_load_error(self):
             }
             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 = {

vllm_ascend/ascend_config.py

@@ -70,6 +70,7 @@ class TorchairGraphConfig:
 
     def __init__(self, torchair_graph_config):
         self.enabled = torchair_graph_config.get("enabled", False)
+        self.mode = torchair_graph_config.get("mode", '')
         self.use_cached_graph = torchair_graph_config.get(
             "use_cached_graph", False)
         self.graph_batch_sizes = torchair_graph_config.get(
@@ -91,6 +92,9 @@ def __init__(self, torchair_graph_config):
                 "graph_batch_sizes_init is only valid when graph_batch_sizes is empty"
             )
         if not self.enabled:
+            if self.mode:
+                raise RuntimeError(
+                    "mode is valid only when Torchair graph mode is enabled")
             if self.use_cached_graph:
                 raise RuntimeError(
                     "use_cached_graph is valid only when Torchair graph mode is enabled"

vllm_ascend/torchair/torchair_model_runner.py

@@ -324,6 +324,8 @@ def _get_torchair_lazy_compiled_model(self, batch_size: int):
             communication_adaptation_310p()
 
         config = torchair.CompilerConfig()
+        if get_ascend_config().torchair_graph_config.mode:
+            config.mode = get_ascend_config().torchair_graph_config.mode
         config.experimental_config.frozen_parameter = True
         # enabling tiling_schedule_optimize on 300I Duo has some bugs, so we have to
         # disable it on 300I Duo platform now.