refactor: compute workers based on cpu

swarms/utils/vllm_wrapper.py

@@ -133,12 +133,11 @@ class VLLMWrapper:
         Returns:
             List[str]: List of model responses.
         """
-        # Fetch 95% of the available CPU cores
+        # Calculate the worker count based on 95% of available CPU cores
-        num_cores = os.cpu_count()
+        num_workers = max(1, int((os.cpu_count() or 1) * 0.95))
-        num_workers = int(num_cores * 0.95)
+        with concurrent.futures.ThreadPoolExecutor(
-        with concurrent.futures.ThreadPoolExecutor(
+            max_workers=num_workers
-            max_workers=num_workers
+        ) as executor:
-        ) as executor:
             futures = [
                 executor.submit(self.run, task) for task in tasks
             ]