[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/spec_decode/eagle_proposer.py

@@ -1,4 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
+from contextlib import nullcontext
 from typing import Optional
 
 import numpy as np
@@ -7,7 +8,9 @@ import torch.nn as nn
 import torch.nn.functional as F
 from vllm.config import (CompilationMode, CUDAGraphMode, VllmConfig,
                          get_layers_from_vllm_config)
-from vllm.distributed.parallel_state import get_pp_group
+from vllm.distributed.parallel_state import (get_pp_group, get_world_group,
+                                             init_model_parallel_group,
+                                             patch_tensor_parallel_group)
 from vllm.forward_context import get_forward_context
 from vllm.logger import logger
 from vllm.model_executor.layers.attention_layer_base import AttentionLayerBase
@@ -93,6 +96,27 @@ class EagleProposer(VllmEagleProposer):
 
         self.use_sparse = hasattr(vllm_config.model_config.hf_text_config,
                                   "index_topk")
+        # NOTE:
+        # `draft_tensor_parallel_size` does not take effect for Eagle:
+        # the draft model uses the same TP size as the target model in practice.
+        # so we applied this patch to set tp=1 of draft model separately.
+        # Due to verification of `_verify_and_get_draft_tp` in vllm,
+        # the value of `draft_tensor_parallel_size` here will either be 1 separately
+        # or the same as target model.
+        # TODO(zhaomingyu13): If we want to adapt to the case where draft model tp
+        # is not 1 and differs from target model, this part should be rewritten.
+        if (vllm_config.parallel_config.tensor_parallel_size
+                != self.speculative_config.draft_tensor_parallel_size):
+            tp_group = init_model_parallel_group(
+                [[get_world_group().rank]],
+                get_world_group().rank,
+                torch.distributed.get_backend(get_world_group().device_group),
+                use_message_queue_broadcaster=True,
+                group_name="tp",
+            )
+            self.tp_group_context = patch_tensor_parallel_group(tp_group)
+        else:
+            self.tp_group_context = nullcontext()
 
     def load_model(self, model: nn.Module) -> None:
         target_attn_layer_names = set(