【Feature】refactor npu_modelrunner for profile_run (#4993)

### What this PR does / why we need it?
(1)refactor npu_model_runner for profile_run
(2) move _select_moe_comm_method to ascend_forward_context
(3) delete _init_model_kwargs in npu_model_runner

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
Na
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: zhenwenqi2024 <zhenwenqi_2022@qq.com>
Signed-off-by: zhenwenqi2024 <155598497+zhenwenqi2024@users.noreply.github.com>

tests/ut/spec_decode/test_eagle_proposer.py

@@ -165,7 +165,6 @@ class TestEagleProposerDummyRun(TestBase):
         self.vllm_config.speculative_config = MagicMock()
         self.device = torch.device("cpu")
         self.runner = MagicMock()
-        self.runner._select_moe_comm_method.return_value = "alltoall"
 
         self.vllm_config.cache_config.block_size = 16
         self.vllm_config.scheduler_config.max_num_batched_tokens = 1024
@@ -192,8 +191,6 @@ class TestEagleProposerDummyRun(TestBase):
     def test_dummy_run_with_prefill(self, mock_context):
         mock_context.return_value.__enter__.return_value = None
         self.proposer.dummy_run(num_tokens=64, with_prefill=True, num_reqs=4)
-
-        self.runner._select_moe_comm_method.assert_called_with(64)
         self.proposer.model.assert_called_once()
 
 

tests/ut/spec_decode/test_mtp_proposer.py

@@ -158,7 +158,6 @@ class TestMtpProposer:
         proposer.model = MagicMock()
         proposer.enable_shared_expert_dp = False
         runner._sync_metadata_across_dp.return_value = (8, 8, False)
-        runner._select_moe_comm_method.return_value = "alltoall"
 
         mock_get_forward_context = MagicMock()
         mock_get_forward_context.cudagraph_runtime_mode = None
@@ -168,7 +167,6 @@ class TestMtpProposer:
 
         # Verify
         runner._sync_metadata_across_dp.assert_called_once()
-        runner._select_moe_comm_method.assert_called_once()
         mock_set_context.assert_called()
 
         # Check that model was called correct number of times
@@ -187,7 +185,6 @@ class TestMtpProposer:
         proposer.enable_shared_expert_dp = False
         proposer.model = MagicMock()
         runner._sync_metadata_across_dp.return_value = (8, 8, False)
-        runner._select_moe_comm_method.return_value = "alltoall"
         runner.attn_groups = []
 
         mock_get_forward_context = MagicMock()
@@ -200,7 +197,6 @@ class TestMtpProposer:
 
         # Verify
         runner._sync_metadata_across_dp.assert_called_once()
-        runner._select_moe_comm_method.assert_called_once()
         mock_set_context.assert_called()
 
         # Check that model was called correct number of times