Logging change

src/handler.py

@@ -20,7 +20,7 @@ engine_args = AsyncEngineArgs(
     tokenizer=config.tokenizer,
     tensor_parallel_size=config.num_gpu_shard,
     dtype=config.dtype,
-    disable_log_stats=True,
+    disable_log_stats=config.disable_log_stats,
     quantization=config.quantization,
     gpu_memory_utilization=0.97,
 )

src/utils.py

@@ -9,6 +9,7 @@ class EngineConfig:
         self.use_full_metrics = os.getenv('USE_FULL_METRICS', 'True') == 'True'
         self.quantization = os.getenv('QUANTIZATION', None)
         self.dtype = "auto" if str(self.quantization).lower() not in ['squeezellm', 'awq'] else "half"
+        self.disable_log_stats = os.getenv('DISABLE_LOG_STATS', 'False') == 'True'
         if make_dirs and not os.path.exists(self.model_base_path):
             os.makedirs(self.model_base_path)