[BugFix] Support setting tp=1 for the Eagle draft model to take effect (#6097)

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
No
```python
from vllm import LLM, SamplingParams

def main():
    prompts = [
        "The future of AI is",
    ]
    sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
    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,
            },
        )
    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}")
```

Fixes vllm-project/vllm#31345

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

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

### How was this patch tested?

- vLLM version: v0.13.0
- vLLM main:
d68209402d

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

tests/e2e/singlecard/spec_decode/test_mtp_eagle_correctness.py

@@ -23,6 +23,7 @@
 from __future__ import annotations
 
 import os
+from typing import Union
 
 import pytest
 from vllm import SamplingParams
@@ -124,11 +125,11 @@ def test_deepseek_mtp_correctness(model_name: str, num_speculative_tokens: int,
 @pytest.mark.parametrize("method", ["eagle", "eagle3"])
 @pytest.mark.parametrize("disable_padded_drafter_batch", [True, False])
 @pytest.mark.parametrize("async_scheduling", [True, False])
-def test_llama_qwen3_eagle_correctness(model_name: str, model_name_main: str,
-                                       num_speculative_tokens: int,
-                                       method: str,
-                                       disable_padded_drafter_batch: bool,
-                                       async_scheduling: bool):
+@pytest.mark.parametrize("draft_tensor_parallel_size", [None, 1])
+def test_llama_qwen3_eagle_correctness(
+        model_name: str, model_name_main: str, num_speculative_tokens: int,
+        method: str, disable_padded_drafter_batch: bool,
+        async_scheduling: bool, draft_tensor_parallel_size: Union[None, int]):
 
     example_prompts = [
         "Hello, my name is",
@@ -163,6 +164,8 @@ def test_llama_qwen3_eagle_correctness(model_name: str, model_name_main: str,
                         "method": method,
                         "model": model_name,
                         "num_speculative_tokens": num_speculative_tokens,
+                        "draft_tensor_parallel_size":
+                        draft_tensor_parallel_size,
                         "max_model_len": 128,
                         "draft_vocab_size": 128256,
                     },

tests/e2e/singlecard/spec_decode/test_v1_spec_decode.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 import math
 import os
 import random
-from typing import Any
+from typing import Any, Union
 
 import pytest
 from transformers import AutoTokenizer
@@ -267,9 +267,11 @@ def test_suffix_acceptance(
 
 
 @pytest.mark.parametrize("use_eagle3", [True], ids=["eagle3"])
+@pytest.mark.parametrize("draft_tensor_parallel_size", [None, 1])
 def test_eagle_logprobs(
     model_name: str,
     use_eagle3: bool,
+    draft_tensor_parallel_size: Union[None, int],
 ):
     prompt = {"role": "user", "content": "Hello world " * 10}
     sampling_params = SamplingParams(temperature=0,
@@ -296,6 +298,7 @@ def test_eagle_logprobs(
                 "method": "eagle3" if use_eagle3 else "eagle",
                 "model": spec_model_name,
                 "num_speculative_tokens": 2,
+                "draft_tensor_parallel_size": draft_tensor_parallel_size,
                 "max_model_len": 128,
             },
             max_model_len=128,
@@ -321,11 +324,13 @@ def test_eagle_logprobs(
 
 @pytest.mark.parametrize("method", MODELS.keys())
 @pytest.mark.parametrize("num_speculative_tokens", [3])
+@pytest.mark.parametrize("draft_tensor_parallel_size", [None, 1])
 @pytest.mark.parametrize("disable_padded_drafter_batch", [True, False])
 @pytest.mark.parametrize("async_scheduling", [True, False])
 def test_llama_qwen_eagle_acceptance(
     method: str,
     num_speculative_tokens: int,
+    draft_tensor_parallel_size: Union[None, int],
     disable_padded_drafter_batch: bool,
     async_scheduling: bool,
 ):
@@ -376,6 +381,7 @@ def test_llama_qwen_eagle_acceptance(
     speculative_config = {
         "method": method,
         "num_speculative_tokens": num_speculative_tokens,
+        "draft_tensor_parallel_size": draft_tensor_parallel_size,
         "disable_padded_drafter_batch": disable_padded_drafter_batch,
         "model": spec_model_name,
     }