[bugfix] fix grpoargs check server_base_url (#5483)

* fix server check

* fix

Author: hjh0119

swift/llm/argument/rlhf_args.py

@@ -213,10 +213,10 @@ def _init_grpo(self):
                     logger.info(f'Auto-configured soft_max_length = max_completion_length {self.max_completion_length}')
             if self.use_vllm:
                 # set vllm mode
-                if self.vllm_server_host is not None:
+                if self.vllm_server_host is not None or self.vllm_server_base_url is not None:
                     if self.vllm_mode != 'server':
                         self.vllm_mode = 'server'
-                        logger.warning('set vllm_mode to `server` since vllm_server_host is provided')
+                        logger.warning('set vllm_mode to `server` since vllm server host/base_url is provided')
                 else:
                     if self.vllm_mode != 'colocate':
                         self.vllm_mode = 'colocate'
@@ -253,7 +253,7 @@ def _init_rm(self):
             self.num_labels = 1
 
     def _init_external_vllm(self):
-        if self.rlhf_type != 'grpo' or self.vllm_server_host is None:
+        if self.rlhf_type != 'grpo' or (self.vllm_server_host is None and self.vllm_server_base_url is None):
             return
         from swift.trainers.rlhf_trainer.vllm_client import VLLMClient
         if is_master():
@@ -313,7 +313,7 @@ def _check_grpo(self):
             assert is_liger_kernel_available(), (
                 'Please install/update liger-kernel by running: pip install -U liger-kernel')
         if self.vllm_mode == 'server':
-            assert not self.use_vllm or self.vllm_server_host is not None
+            assert not self.use_vllm or self.vllm_server_host is not None or self.vllm_server_base_url is not None
 
         if self.async_generate:
             assert self.vllm_mode == 'server', 'async generate require vllm_mode == server, '