diff --git a/llm_server/llm/openai/oai_to_vllm.py b/llm_server/llm/openai/oai_to_vllm.py
index 35c9f30..1490933 100644
--- a/llm_server/llm/openai/oai_to_vllm.py
+++ b/llm_server/llm/openai/oai_to_vllm.py
@@ -27,6 +27,8 @@ def oai_to_vllm(request_json_body, stop_hashes: bool, mode):
     if mode == 'vllm' and request_json_body.get('top_p') == 0:
         request_json_body['top_p'] = 0.01
 
+    request_json_body['max_tokens'] = min(max(request_json_body.get('max_new_tokens', 0), request_json_body.get('max_tokens', 0)), opts.max_new_tokens)
+
     return request_json_body
 
 
diff --git a/llm_server/routes/request_handler.py b/llm_server/routes/request_handler.py
index a048df7..dd8326b 100644
--- a/llm_server/routes/request_handler.py
+++ b/llm_server/routes/request_handler.py
@@ -37,6 +37,7 @@ class RequestHandler:
         self.parameters = None
         self.used = False
 
+        self.selected_model = selected_model
         self.backend_url = get_a_cluster_backend(selected_model)
         self.cluster_backend_info = cluster_config.get_backend(self.backend_url)