[CI] speed up ut (#4901)

avoid model download to speed up ut test. 

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

vllm_ascend/eplb/core/policy/policy_dynamic_ep_v2.py

@@ -4,6 +4,7 @@ from abc import abstractmethod
 from collections import defaultdict
 
 import numpy as np
+from vllm.logger import logger
 
 
 class DynamicConfig:
@@ -69,21 +70,21 @@ class DynamicEplbV2(EplbPolicy):
     @staticmethod
     def safe_divide(a, b):
         if b == 0:
-            print("Division by zero is not allowed")
+            logger.info("Division by zero is not allowed")
             return 0
         return a / b
 
     @staticmethod
     def safe_exact_divide(a, b):
         if b == 0:
-            print("Division by zero is not allowed")
+            logger.info("Division by zero is not allowed")
             return 0
         return a // b
 
     @staticmethod
     def safe_mod(a, b):
         if b == 0:
-            print("Division by zero is not allowed")
+            logger.info("Division by zero is not allowed")
             return 0
         return a % b
 

vllm_ascend/model_loader/netloader/interaction/elastic.py

@@ -60,7 +60,7 @@ class ElasticClient:
                 ip, port_str = source.split(':')
                 port = int(port_str)
             except Exception as e:
-                logger.error(f"IP format error: {source}, detail: {e}")
+                logger.info(f"IP format error: {source}, detail: {e}")
                 continue
 
             self.server_addr = ip

vllm_ascend/model_loader/netloader/load.py

@@ -80,5 +80,5 @@ def elastic_load(
                 time.perf_counter() - t0))
             return model_loaded
     except Exception as e:
-        logger.error(f"elastic_load error: {e}")
+        logger.info(f"elastic_load error: {e}")
         return None