[fix]: use env varibles to enable dbo model

Signed-off-by: zhuohuan <zxdu1997@gmail.com>

Author: zxdukki

examples/offline_dualbatch_overlap_npu.py

@@ -15,24 +15,20 @@
 
 def main():
     # Create an LLM.
-    llm = LLM(
-        model="deepseek-ai/DeepSeek-V3-Lite-base-latest-w8a8-dynamic",
-        hf_overrides={
-            "architectures": ["DeepseekDBOForCausalLM"],
-        },  # override the model arch to run the dbo model
-        enforce_eager=True,
-        tensor_parallel_size=2,
-        max_model_len=4096,
-        trust_remote_code=True,
-        additional_config={
-            "torchair_graph_config": {
-                "enabled": False
-            },
-            "ascend_scheduler_config": {
-                "enabled": True
-            },
-            "expert_tensor_parallel_size": 1
-        })
+    llm = LLM(model="deepseek-ai/DeepSeek-V3-Lite-base-latest-w8a8-dynamic",
+              enforce_eager=True,
+              tensor_parallel_size=2,
+              max_model_len=4096,
+              trust_remote_code=True,
+              additional_config={
+                  "torchair_graph_config": {
+                      "enabled": False
+                  },
+                  "ascend_scheduler_config": {
+                      "enabled": True
+                  },
+                  "expert_tensor_parallel_size": 1
+              })
 
     # Generate texts from the prompts. The output is a list of RequestOutput
     # objects that contain the prompt, generated text, and other information.

tests/multicard/test_offline_inference_distributed.py

@@ -93,8 +93,5 @@ def test_models_distributed_DeepSeek_dbo():
             dtype=dtype,
             tensor_parallel_size=4,
             distributed_executor_backend="mp",
-            hf_overrides={
-                "architectures": ["DeepseekDBOForCausalLM"],
-            }  # override the model arch to the dbo version
     ) as vllm_model:
         vllm_model.generate(example_prompts, sampling_params)

vllm_ascend/models/__init__.py

@@ -1,5 +1,7 @@
 from vllm import ModelRegistry
 
+import vllm_ascend.envs as envs
+
 
 def register_model():
     from .deepseek_dbo import CustomDeepseekDBOForCausalLM  # noqa: F401
@@ -23,9 +25,14 @@ def register_model():
         "vllm_ascend.models.qwen2_5_vl:AscendQwen2_5_VLForConditionalGeneration"
     )
 
-    ModelRegistry.register_model(
-        "DeepseekV2ForCausalLM",
-        "vllm_ascend.models.deepseek_v2:CustomDeepseekV2ForCausalLM")
+    if envs.VLLM_ASCEND_ENABLE_DBO:
+        ModelRegistry.register_model(
+            "DeepseekV2ForCausalLM",
+            "vllm_ascend.models.deepseek_dbo:CustomDeepseekDBOForCausalLM")
+    else:
+        ModelRegistry.register_model(
+            "DeepseekV2ForCausalLM",
+            "vllm_ascend.models.deepseek_v2:CustomDeepseekV2ForCausalLM")
 
     ModelRegistry.register_model(
         "DeepseekV3ForCausalLM",
@@ -34,7 +41,3 @@ def register_model():
     ModelRegistry.register_model(
         "Qwen3MoeForCausalLM",
         "vllm_ascend.models.qwen3_moe:CustomQwen3MoeForCausalLM")
-
-    ModelRegistry.register_model(
-        "DeepseekDBOForCausalLM",
-        "vllm_ascend.models.deepseek_dbo:CustomDeepseekDBOForCausalLM")