[bugfix] asyncscheduler bug fix (#4968)

### What this PR does / why we need it?
now vllm-ascend uses AsyncGPUModelRunnerOutput
,AsyncNPUModelRunnerOutput before is outdated, so we should fix it

- vLLM version: v0.12.0
- vLLM main:
vllm-project/vllm@ad32e3e

---------

Signed-off-by: zhenwenqi2024 <[email protected]>

Author: zhenwenqi2024

.github/workflows/_e2e_test.yaml

@@ -93,6 +93,7 @@ jobs:
           pytest -sv tests/e2e/singlecard/test_completion_with_prompt_embeds.py
           pytest -sv tests/e2e/singlecard/test_aclgraph_accuracy.py
           pytest -sv tests/e2e/singlecard/test_aclgraph_mem.py
+          pytest -sv tests/e2e/singlecard/test_async_scheduling.py
           pytest -sv tests/e2e/singlecard/test_camem.py
           pytest -sv tests/e2e/singlecard/test_guided_decoding.py
           # torch 2.8 doesn't work with lora, fix me

tests/e2e/singlecard/test_async_scheduling.py

@@ -17,8 +17,12 @@
 
 first_prompt = ("The following numbers of the sequence " +
                 ", ".join(str(i) for i in range(10)) + " are:")
-example_prompts = [first_prompt, "In one word, the capital of France is "
-                   ] + [f"Tell me about the number {i}: " for i in range(32)]
+example_prompts = [
+    "Hello, my name is",
+    "The president of the United States is",
+    "The capital of France is",
+    "The future of AI is",
+]
 
 default_params = dict(
     temperature=0.0,  # greedy

vllm_ascend/worker/model_runner_v1.py

@@ -1875,8 +1875,10 @@ def propose_draft_token_ids(sampled_token_ids):
         return AsyncGPUModelRunnerOutput(
             model_runner_output=model_runner_output,
             sampled_token_ids=sampled_token_ids,
+            logprobs_tensors=sampler_output.logprobs_tensors,
             invalid_req_indices=invalid_req_indices,
             async_output_copy_stream=self.async_output_copy_stream,
+            vocab_size=self.input_batch.vocab_size,
         )
 
     def _build_dummy_attn_metadata(
@@ -3472,7 +3474,7 @@ def __init__(self, *args, **kwargs) -> None:
 
     try:
         # replace cuda APIs with xpu APIs, this should work by default
-        torch.cuda.Event = _EventPlaceholder
+        torch.cuda.Event = torch.npu.Event
         torch.cuda.Stream = torch.npu.Stream
         torch.cuda.default_stream = torch.npu.default_stream
         torch.cuda.current_stream = torch.npu.current_stream