[Main2Main] Upgrade vllm commit to 0105 (#5595)

### What this PR does / why we need it?

Upgrade vllm commit to 0105 (8be6432bdaf6275664d857b1e5e9bf8ed1ce299e)

1. Remove `maybe_padded_num_tokens` arg in `model_runner_v1.py` since
https://github.com/vllm-project/vllm/pull/31517 deleted unused arg

2. Remove dense `Qwen/Qwen3-0.6B` in
`tests/e2e/multicard/test_aclgraph_capture_replay.py` and
`tests/e2e/multicard/test_data_parallel.py` due to
https://github.com/vllm-project/vllm/pull/30739
where offline data parallel mode will not be supported/useful for dense
models

3. Adapt `vllm_ascend/worker/worker.py` due to
https://github.com/vllm-project/vllm/pull/31584

4. Adapt `self.block_size` calling due to
https://github.com/vllm-project/vllm/pull/31540

5. Modify `test_mla_v1.py` due to
https://github.com/vllm-project/vllm/pull/28454 , which refactorred
`get_head_size()`

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.13.0
- vLLM main:
7157596103

Signed-off-by: wjunLu <wjunlu217@gmail.com>

vllm_ascend/attention/attention_v1.py

@@ -197,6 +197,7 @@ class AscendAttentionMetadataBuilder(AttentionMetadataBuilder[AscendMetadata]):
         vllm_config: VllmConfig,
         device: torch.device,
     ):
+        super().__init__(kv_cache_spec, layer_names, vllm_config, device)
         self.vllm_config = vllm_config
         self.model_config = vllm_config.model_config
         self.compilation_config = vllm_config.compilation_config

vllm_ascend/spec_decode/eagle_proposer.py

@@ -136,6 +136,7 @@ class EagleProposer(VllmEagleProposer):
         draft_attn_layer_names = draft_attn_layer_names - draft_indexer_layer_names
         assert len(draft_attn_layer_names) == 1
         self.attn_layer_name = list(draft_attn_layer_names)
+        self.attn_layer_names = self.attn_layer_name
 
         # share embed_tokens with the target model if needed
         if get_pp_group().world_size == 1:
@@ -442,14 +443,19 @@ class EagleProposer(VllmEagleProposer):
             # For the requests that exceed the max model length, we set the
             # TODO: sequence length to 1 to minimize their overheads in attention.
 
+            if self.attn_metadata_builder is None:
+                attn_metadata_builder = self._get_attention_metadata_builder()
+            else:
+                attn_metadata_builder = self.attn_metadata_builder
+            block_size = attn_metadata_builder.kv_cache_spec.block_size
+
             # Compute the slot mapping.
-            block_numbers = (clamped_positions // self.block_size)
+            block_numbers = (clamped_positions // block_size)
             block_ids = attn_metadata.block_tables.gather(
                 dim=1, index=block_numbers.view(-1, 1))
             block_ids = block_ids.view(-1)
-            slot_mapping_tmp = (
-                block_ids * self.vllm_config.cache_config.block_size +
-                clamped_positions % self.block_size)
+            slot_mapping_tmp = (block_ids * block_size +
+                                clamped_positions % block_size)
 
             # Mask out the slot mappings that exceed the max model length.
             # Otherwise, the KV cache will be inadvertently updated with the

vllm_ascend/worker/model_runner_v1.py

@@ -107,7 +107,7 @@ from vllm_ascend.spec_decode.mtp_proposer import MtpProposer
 from vllm_ascend.utils import (AscendDeviceType, ProfileExecuteDuration,
                                enable_sp, get_ascend_device_type, is_moe_model,
                                lmhead_tp_enable, maybe_trans_nz,
-                               set_weight_prefetch_method)
+                               set_weight_prefetch_method, vllm_version_is)
 from vllm_ascend.worker.npu_input_batch import NPUInputBatch
 from vllm_ascend.worker.pcp_utils import PCPManager
 
@@ -1097,12 +1097,20 @@ class NPUModelRunner(GPUModelRunner):
                                              intermediate_tensors,
                                              inputs_embeds):
         assert self.model is not None
-        hidden_states = self.model(
-            input_ids=input_ids,
-            positions=positions,
-            intermediate_tensors=intermediate_tensors,
-            inputs_embeds=inputs_embeds,
-            **self._init_model_kwargs(maybe_padded_num_tokens))
+        if vllm_version_is('0.13.0'):
+            hidden_states = self.model(
+                input_ids=input_ids,
+                positions=positions,
+                intermediate_tensors=intermediate_tensors,
+                inputs_embeds=inputs_embeds,
+                **self._init_model_kwargs(maybe_padded_num_tokens))
+        else:
+            hidden_states = self.model(
+                input_ids=input_ids,
+                positions=positions,
+                intermediate_tensors=intermediate_tensors,
+                inputs_embeds=inputs_embeds,
+                **self._init_model_kwargs())
 
         forward_context = get_forward_context()
         if forward_context.cudagraph_runtime_mode == CUDAGraphMode.FULL \
@@ -1548,10 +1556,16 @@ class NPUModelRunner(GPUModelRunner):
                 logits = None
             else:
                 if self.input_batch.pooling_params:
-                    pool_output = self._pool(
-                        hidden_states,
-                        scheduler_output.total_num_scheduled_tokens,
-                        num_scheduled_tokens_np)
+                    if vllm_version_is('0.13.0'):
+                        pool_output = self._pool(
+                            hidden_states,
+                            scheduler_output.total_num_scheduled_tokens,
+                            num_scheduled_tokens_np)
+                    else:
+                        pool_output = self._pool(
+                            hidden_states,
+                            scheduler_output.total_num_scheduled_tokens,
+                            num_scheduled_tokens_np, kv_connector_output)
                     if self.debugger is not None:
                         self.debugger.stop()
                         self.debugger.step()

vllm_ascend/worker/worker.py

@@ -299,7 +299,7 @@ class NPUWorker(WorkerBase):
     def execute_model(
         self,
         scheduler_output: "SchedulerOutput",
-    ) -> ModelRunnerOutput | None:
+    ) -> ModelRunnerOutput | AsyncModelRunnerOutput | None:
         # enable msMonitor to monitor the performance of vllm-ascend
         if envs_ascend.MSMONITOR_USE_DAEMON:
             dp.step()
@@ -318,7 +318,8 @@ class NPUWorker(WorkerBase):
 
         output = self.model_runner.execute_model(scheduler_output,
                                                  intermediate_tensors)
-        if isinstance(output, (ModelRunnerOutput, NoneType)):
+        if isinstance(output,
+                      (ModelRunnerOutput, AsyncModelRunnerOutput, NoneType)):
             return output
 
         assert isinstance(output, IntermediateTensors)