[Iluvatar] Fix FD launch error when specifing CUDA_VISBLE_DEVICE

Author: wuyujiji

docs/get_started/installation/iluvatar_gpu.md

@@ -480,7 +480,7 @@ python3 -m fastdeploy.entrypoints.openai.api_server \
        --cache-queue-port 55660 \
        --max-model-len 16384 \
        --max-num-batched-tokens 16384 \
-       --max-num-seqs 32 \
+       --max-num-seqs 64 \
        --workers 2 \
        --block-size 16
 ```

docs/zh/get_started/installation/iluvatar_gpu.md

@@ -480,7 +480,7 @@ python3 -m fastdeploy.entrypoints.openai.api_server \
        --cache-queue-port 55660 \
        --max-model-len 16384 \
        --max-num-batched-tokens 16384 \
-       --max-num-seqs 32 \
+       --max-num-seqs 64 \
        --workers 2 \
        --block-size 16
 ```

fastdeploy/engine/engine.py

@@ -577,6 +577,10 @@ def _start_worker_service(self):
         if self.cfg.structured_outputs_config.logits_processors is not None:
             arguments += f" --logits-processors {' '.join(self.cfg.structured_outputs_config.logits_processors)}"
 
+        # TODO (iluvatar): remove aftet paddle fix launch error
+        if current_platform.is_iluvatar() and "CUDA_VISIBLE_DEVICES" in os.environ:
+            arguments = arguments.replace(f"--devices {self.cfg.parallel_config.device_ids}", "")
+
         worker_store_true_flag = {
             "enable_expert_parallel": self.cfg.parallel_config.enable_expert_parallel,
             "enable_chunked_moe": self.cfg.parallel_config.enable_chunked_moe,