print out llm metrics, make metrics more verbose

src/handler.py

@@ -15,7 +15,7 @@ MODEL_NAME = os.environ.get('MODEL_NAME')
 MODEL_BASE_PATH = os.environ.get('MODEL_BASE_PATH', '/runpod-volume/')
 STREAMING = os.environ.get('STREAMING', False) == 'True'
 TOKENIZER = os.environ.get('TOKENIZER', None)
-USE_FULL_METRICS = os.environ.get('USE_FULL_METRICS', False)
+USE_FULL_METRICS = os.environ.get('USE_FULL_METRICS', True)
 
 if not MODEL_NAME:
     print("Error: The model has not been provided.")

src/metrics.py

@@ -68,4 +68,7 @@ def vllm_log_system_stats(
             'cpu_kv_cache_usage': cpu_cache_usage, # percentage
         }
 
+        # Print metrics
+        print(metrics)
+
         self.metrics = metrics