feat: support data parallel for deepseek (#1012)

### What this PR does / why we need it?
feat: support data parallel for deepseek

### Does this PR introduce _any_ user-facing change?
Yes, support dp for deepseek

### How was this patch tested?

```
export VLLM_ENABLE_MC2=0
export VLLM_USE_V1=1
export TASK_QUEUE_ENABLE=1

source /usr/local/Ascend/ascend-toolkit/set_env.sh
source /usr/local/Ascend/nnal/atb/set_env.sh

nohup python -m vllm.entrypoints.openai.api_server
--model=/path/to/DeepSeek-R1-W8A8 \
    --quantization ascend \
    --served-model-name auto \
    --trust-remote-code \
    --distributed-executor-backend=mp \
    --port 8006 \
    -tp=8 \
    -dp=2 \
    --max-num-seqs 24 \
    --max-model-len 4096 \
    --max-num-batched-tokens 4096 \
    --block-size 128 \
    -O 0 \
    --no-enable-prefix-caching \
--additional-config
'{"torchair_graph_batch_sizes":[24],"expert_tensor_parallel_size":16,"ascend_scheduler_config":{},"enable_graph_mode":true}'
\
    --gpu-memory-utilization 0.95 &> run.log &
disown
```

Signed-off-by: boying <897013703@qq.com>

vllm_ascend/quantization/w8a8_dynamic.py

@@ -20,6 +20,7 @@ from typing import Any, Callable, Dict, Optional
 import torch
 import torch.distributed as dist
 import torch_npu
+from vllm.config import get_current_vllm_config
 from vllm.distributed import GroupCoordinator
 
 import vllm_ascend.envs as envs_ascend
@@ -508,6 +509,12 @@ class AscendW8A8DynamicFusedMoEMethod:
 
         self.ep_group = get_ep_group()
 
+        self.enable_graph_mode = False
+        additional_config = get_current_vllm_config().additional_config
+        if additional_config:
+            self.enable_graph_mode = additional_config.get(
+                "enable_graph_mode", False)
+
         try:
             device_group = self.ep_group.device_group
             # TODO: Try local_rank = ep_group.rank_in_group
@@ -629,7 +636,7 @@ class AscendW8A8DynamicFusedMoEMethod:
                 top_k=top_k,
                 expert_map=expert_map,
                 moe_all_to_all_group_name=self.moe_all_to_all_group_name)
-        elif self.ep_group.world_size == 1:
+        elif self.enable_graph_mode or self.ep_group.world_size == 1:
             return fused_experts(hidden_states=x,
                                  w1=layer.w13_weight,
                                  w1_scale=layer.w13_weight_scale,