From 7762cd6bed7f95fa855144e01aa94e932869fe12 Mon Sep 17 00:00:00 2001 From: sigoden Date: Wed, 8 May 2024 13:46:26 +0800 Subject: refactor: model pass_max_tokens (#493) --- src/serve.rs | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'src/serve.rs') diff --git a/src/serve.rs b/src/serve.rs index 5f748e8..4413c89 100644 --- a/src/serve.rs +++ b/src/serve.rs @@ -93,7 +93,7 @@ impl Server { "id": id, "max_input_tokens": model.max_input_tokens, "max_output_tokens": model.max_output_tokens, - "max_output_tokens?": model.ref_max_output_tokens, + "pass_max_tokens": model.pass_max_tokens, "input_price": model.input_price, "output_price": model.output_price, "supports_vision": model.supports_vision(), @@ -244,7 +244,7 @@ impl Server { let mut client = init_client(&config)?; if max_tokens.is_some() { - client.model_mut().set_max_output_tokens(max_tokens); + client.model_mut().set_max_tokens(max_tokens, true); } let abort = create_abort_signal(); let http_client = client.build_client()?; -- cgit v1.2.3