[BugFix] Support setting tp=1 for the Eagle draft model to take effect (#5519)

### What this PR does / why we need it?
According to the official documentation, the parameter
"draft_tensor_parallel_size": 1 is supposed to be applied to the Eagle3
model. However, based on actual debugging, it was found that the number
of tensor parallelisms (tp) of the Eagle model is consistent with that
of the target model. The setting of tp for the draft model did not take
effect as expected.

**Note:** This feature has not been superimposed and tested with `sp`
and `dp`. It will be adapted later
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
```python
from vllm import LLM, SamplingParams

def main():
    prompts = [
        "The future of AI is",
    ]

    # Create a sampling params object.
    sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
    # Create an LLM.
    llm = LLM(
            model="meta-llama/Llama-3.1-8B-Instruct",
            tensor_parallel_size=4,
            gpu_memory_utilization=0.9,
            enforce_eager=True,
            speculative_config={
                "method": "eagle3",
                "model": "yuhuili/EAGLE3-LLaMA3.1-Instruct-8B"
                "draft_tensor_parallel_size": 1,
                "num_speculative_tokens": 3,
            },
        )

    # Generate texts from the prompts.
    outputs = llm.generate(prompts, sampling_params)
    print(f"Outputs: {outputs}")
    for output in outputs:
        prompt = output.prompt
        generated_text = output.outputs[0].text
        print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
- vLLM version: v0.13.0
- vLLM main:
45c1ca1ca1

Fixes vllm-project/vllm#31345

Signed-off-by: zhaomingyu <zhaomingyu13@h-partners.com>
Co-authored-by: drslark <slarksblood@qq.com>

vllm_ascend/worker/model_runner_v1.py

@@ -165,6 +165,10 @@ def graph_capture(device: torch.device):
         yield graph_capture_context
 
 
+def get_tp_context(drafter):
+    return getattr(drafter, "tp_group_context", nullcontext())
+
+
 class ExecuteModelState(NamedTuple):
     """Ephemeral cached state transferred between execute_model() and
     sample_tokens(), after execute_model() returns None."""
@@ -2320,7 +2324,8 @@ class NPUModelRunner(GPUModelRunner):
                 model_register(self.model, self.model_config)
             if self.drafter:
                 logger.info("Loading drafter model...")
-                self.drafter.load_model(self.model)
+                with get_tp_context(self.drafter):
+                    self.drafter.load_model(self.model)
                 if self.use_aux_hidden_state_outputs:
                     self.model.set_aux_hidden_state_layers(
                         self.model.get_eagle3_aux_hidden_state_layers())
@@ -2696,11 +2701,15 @@ class NPUModelRunner(GPUModelRunner):
         kernel_block_sizes = []
         for kv_cache_group_id, kv_cache_group in enumerate(
                 kv_cache_config.kv_cache_groups):
-
-            if isinstance(kv_cache_group.kv_cache_spec,
-                          EncoderOnlyAttentionSpec):
+            kv_cache_spec = kv_cache_group.kv_cache_spec
+            if isinstance(kv_cache_spec, UniformTypeKVCacheSpecs):
+                # All layers in the UniformTypeKVCacheSpecs have the same type,
+                # Pick an arbitrary one to dispatch.
+                kv_cache_spec = next(
+                    iter(kv_cache_spec.kv_cache_specs.values()))
+            if isinstance(kv_cache_spec, EncoderOnlyAttentionSpec):
                 continue
-            elif isinstance(kv_cache_group.kv_cache_spec, AttentionSpec):
+            elif isinstance(kv_cache_spec, AttentionSpec):
                 # This is an attention backend that supports virtual
                 # block splitting. Get the supported block sizes from
                 # the backend.