From 77db34a6a7669605623a7dcbe09b2c856a8b1448 Mon Sep 17 00:00:00 2001 From: Cyberes Date: Wed, 4 Oct 2023 12:59:19 -0600 Subject: [PATCH] g --- llm_server/routes/request_handler.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/llm_server/routes/request_handler.py b/llm_server/routes/request_handler.py index d0990ea..1361fb4 100644 --- a/llm_server/routes/request_handler.py +++ b/llm_server/routes/request_handler.py @@ -36,11 +36,13 @@ class RequestHandler: self.token_priority, self.token_simultaneous_ip = get_token_ratelimit(self.token) self.backend_url = get_a_cluster_backend(selected_model) self.cluster_backend_info = cluster_config.get_backend(self.backend_url) - self.selected_model = self.cluster_backend_info['model'] if not self.cluster_backend_info.get('mode'): print('keyerror: mode -', selected_model, self.backend_url, self.cluster_backend_info) + if not self.cluster_backend_info.get('model'): + print('keyerror: mode -', selected_model, self.backend_url, self.cluster_backend_info) + self.selected_model = self.cluster_backend_info['model'] self.backend = get_backend_handler(self.cluster_backend_info['mode'], self.backend_url) self.parameters = None self.used = False