refactor: model pass_max_tokens (#493)

assets/arena.html

@@ -746,8 +746,8 @@
             messages: messages,
             stream: true,
           };
-          const { max_output_token, need_max_tokens } = retrieveModel(this.models, chat.model_id);
-          if (!body["max_tokens"] && need_max_tokens) {
+          const { max_output_token, pass_max_tokens } = retrieveModel(this.models, chat.model_id);
+          if (!body["max_tokens"] && pass_max_tokens) {
             body["max_tokens"] = max_output_token;
           };
           return body;
@@ -819,14 +819,14 @@
     function retrieveModel(models, id) {
       const model = models.find(model => model.id === id);
       if (!model) return {};
-      const max_output_token = model.max_output_tokens || model["max_output_tokens?"] || null;
-      const need_max_tokens = !!model.max_output_tokens;
+      const max_output_token = model.max_output_tokens;
       const supports_vision = !!model.supports_vision;
+      const pass_max_tokens = !!model.pass_max_tokens;
       return {
         id,
         max_output_token,
-        need_max_tokens,
         supports_vision,
+        pass_max_tokens,
       }
     }
 

assets/playground.html

@@ -939,8 +939,8 @@
               body[body_key || setting_key] = this.settings[setting_key];
             }
           });
-          const { max_output_token, need_max_tokens } = this.currentModel;
-          if (!body["max_tokens"] && need_max_tokens) {
+          const { max_output_token, pass_max_tokens } = this.currentModel;
+          if (!body["max_tokens"] && pass_max_tokens) {
             body["max_tokens"] = max_output_token;
           };
           return body;
@@ -1013,14 +1013,14 @@
     function retrieveModel(models, id) {
       const model = models.find(model => model.id === id);
       if (!model) return {};
-      const max_output_token = model.max_output_tokens || model["max_output_tokens?"] || null;
-      const need_max_tokens = !!model.max_output_tokens;
+      const max_output_token = model.max_output_tokens;
       const supports_vision = !!model.supports_vision;
+      const pass_max_tokens = !!model.pass_max_tokens;
       return {
         id,
         max_output_token,
-        need_max_tokens,
         supports_vision,
+        pass_max_tokens,
       }
     }
 

config.example.yaml

@@ -33,7 +33,6 @@ clients:
   #   models:
   #     - name: xxxx                                  # The model name
   #       max_input_tokens: 100000
-  #       max_output_tokens: 4096
   #       supports_vision: true
   #       extra_fields:                               # Set custom parameters, will merge with the body json
   #          key: value                               

models.yaml

@@ -8,44 +8,45 @@
   models:
     - name: gpt-3.5-turbo
       max_input_tokens: 16385
-      max_output_tokens?: 4096
+      max_output_tokens: 4096
       input_price: 0.5
       output_price: 1.5
     - name: gpt-3.5-turbo-1106
       max_input_tokens: 16385
-      max_output_tokens?: 4096
+      max_output_tokens: 4096
       input_price: 1
       output_price: 2
     - name: gpt-4-turbo
       max_input_tokens: 128000
-      max_output_tokens?: 4096
+      max_output_tokens: 4096
       input_price: 10
       output_price: 30
       supports_vision: true
     - name: gpt-4-turbo-preview
       max_input_tokens: 128000
-      max_output_tokens?: 4096
+      max_output_tokens: 4096
       input_price: 10
       output_price: 30
     - name: gpt-4-1106-preview
       max_input_tokens: 128000
-      max_output_tokens?: 4096
+      max_output_tokens: 4096
       input_price: 10
       output_price: 30
     - name: gpt-4-vision-preview
       max_input_tokens: 128000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 10
       output_price: 30
       supports_vision: true
     - name: gpt-4
       max_input_tokens: 8192
-      max_output_tokens?: 4096
+      max_output_tokens: 4096
       input_price: 30
       output_price: 60
     - name: gpt-4-32k
       max_input_tokens: 32768
-      max_output_tokens?: 4096
+      max_output_tokens: 4096
       input_price: 60
       output_price: 120
 
@@ -59,18 +60,18 @@
   models:
     - name: gemini-1.0-pro-latest
       max_input_tokens: 30720
-      max_output_tokens?: 2048
+      max_output_tokens: 2048
       input_price: 0.5
       output_price: 1.5
     - name: gemini-1.0-pro-vision-latest
       max_input_tokens: 12288
-      max_output_tokens?: 4096
+      max_output_tokens: 4096
       input_price: 0.5
       output_price: 1.5
       supports_vision: true
     - name: gemini-1.5-pro-latest
       max_input_tokens: 1048576
-      max_output_tokens?: 8192
+      max_output_tokens: 8192
       input_price: 7
       output_price: 21
       supports_vision: true
@@ -85,18 +86,21 @@
     - name: claude-3-opus-20240229
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 15
       output_price: 75
       supports_vision: true
     - name: claude-3-sonnet-20240229
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 3
       output_price: 15
       supports_vision: true
     - name: claude-3-haiku-20240307
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 0.25
       output_price: 1.25
       supports_vision: true
@@ -140,12 +144,12 @@
   models:
     - name: command-r
       max_input_tokens: 128000
-      max_output_tokens?: 4000
+      max_output_tokens: 4000
       input_price: 0.5
       output_price: 1.5
     - name: command-r-plus
       max_input_tokens: 128000
-      max_output_tokens?: 4000
+      max_output_tokens: 4000
       input_price: 3
       output_price: 15
 
@@ -159,28 +163,28 @@
   models:
     - name: llama-3-sonar-small-32k-chat
       max_input_tokens: 32768
-      max_output_tokens?: 32768
+      max_output_tokens: 32768
       input_price: 0.2
       output_price: 0.2
     - name: llama-3-sonar-large-32k-chat	
       max_input_tokens: 32768
-      max_output_tokens?: 32768
+      max_output_tokens: 32768
       input_price: 0.6
       output_price: 0.6
 
     - name: llama-3-8b-instruct
       max_input_tokens: 8192
-      max_output_tokens?: 8192
+      max_output_tokens: 8192
       input_price: 0.2
       output_price: 0.2
     - name: llama-3-70b-instruct
       max_input_tokens: 8192
-      max_output_tokens?: 8192
+      max_output_tokens: 8192
       input_price: 1
       output_price: 1
     - name: mixtral-8x7b-instruct
       max_input_tokens: 16384
-      max_output_tokens?: 16384
+      max_output_tokens: 16384
       input_price: 0.6
       output_price: 0.6
 
@@ -195,22 +199,22 @@
   models:
     - name: llama3-8b-8192
       max_input_tokens: 8192
-      max_output_tokens?: 8192
+      max_output_tokens: 8192
       input_price: 0.05
       output_price: 0.10
     - name: llama3-70b-8192
       max_input_tokens: 8192
-      max_output_tokens?: 8192
+      max_output_tokens: 8192
       input_price: 0.59
       output_price: 0.79
     - name: mixtral-8x7b-32768
       max_input_tokens: 32768
-      max_output_tokens?: 32768
+      max_output_tokens: 32768
       input_price: 0.27
       output_price: 0.27
     - name: gemma-7b-it
       max_input_tokens: 8192
-      max_output_tokens?: 8192
+      max_output_tokens: 8192
       input_price: 0.10
       output_price: 0.10
 
@@ -224,18 +228,18 @@
   models:
     - name: gemini-1.0-pro
       max_input_tokens: 24568
-      max_output_tokens?: 8192
+      max_output_tokens: 8192
       input_price: 0.125
       output_price: 0.375
     - name: gemini-1.0-pro-vision
       max_input_tokens: 14336
-      max_output_tokens?: 2048
+      max_output_tokens: 2048
       input_price: 0.125
       output_price: 0.375
       supports_vision: true
     - name: gemini-1.5-pro-preview-0409
       max_input_tokens: 1000000
-      max_output_tokens?: 8192
+      max_output_tokens: 8192
       input_price: 2.5
       output_price: 7.5
       supports_vision: true
@@ -250,18 +254,21 @@
     - name: claude-3-opus@20240229
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 15
       output_price: 75
       supports_vision: true
     - name: claude-3-sonnet@20240229
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 3
       output_price: 15
       supports_vision: true
     - name: claude-3-haiku@20240307
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 0.25
       output_price: 1.25
       supports_vision: true
@@ -277,44 +284,52 @@
     - name: anthropic.claude-3-opus-20240229-v1:0
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 15
       output_price: 75
       supports_vision: true
     - name: anthropic.claude-3-sonnet-20240229-v1:0
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 3
       output_price: 15
       supports_vision: true
     - name: anthropic.claude-3-haiku-20240307-v1:0
       max_input_tokens: 200000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 0.25
       output_price: 1.25
       supports_vision: true
     - name: meta.llama3-8b-instruct-v1:0
       max_input_tokens: 8192
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 0.4
       output_price: 0.6
     - name: meta.llama3-70b-instruct-v1:0
       max_input_tokens: 8192
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 2.65
       output_price: 3.5
     - name: mistral.mistral-7b-instruct-v0:2
       max_input_tokens: 32000
       max_output_tokens: 8192
+      pass_max_tokens: true
       input_price: 0.15
       output_price: 0.2
     - name: mistral.mixtral-8x7b-instruct-v0:1
       max_input_tokens: 32000
       max_output_tokens: 8192
+      pass_max_tokens: true
       input_price: 0.45
       output_price: 0.7
     - name: mistral.mistral-large-2402-v1:0
       max_input_tokens: 32000
       max_output_tokens: 8192
+      pass_max_tokens: true
       input_price: 8
       output_price: 2.4
 
@@ -328,21 +343,27 @@
     - name: '@cf/meta/llama-3-8b-instruct'
       max_input_tokens: 4096
       max_output_tokens: 4096
+      pass_max_tokens: true
     - name: '@cf/mistral/mistral-7b-instruct-v0.2-lora'
       max_input_tokens: 4096
       max_output_tokens: 4096
+      pass_max_tokens: true
     - name: '@cf/google/gemma-7b-it-lora'
       max_input_tokens: 4096
       max_output_tokens: 4096
+      pass_max_tokens: true
     - name: '@cf/qwen/qwen1.5-14b-chat-awq'
       max_input_tokens: 4096
       max_output_tokens: 4096
+      pass_max_tokens: true
     - name: '@hf/thebloke/deepseek-coder-6.7b-instruct-awq'
       max_input_tokens: 4096
       max_output_tokens: 4096
+      pass_max_tokens: true
     - name: '@hf/nexusflow/starling-lm-7b-beta'
       max_input_tokens: 4096
       max_output_tokens: 4096
+      pass_max_tokens: true
 
 - platform: replicate
   # docs:
@@ -354,21 +375,25 @@
     - name: meta/meta-llama-3-70b-instruct
       max_input_tokens: 8192
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 0.65
       output_price: 2.75
     - name: meta/meta-llama-3-8b-instruct
       max_input_tokens: 8192
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 0.05
       output_price: 0.25
     - name: mistralai/mistral-7b-instruct-v0.2
       max_input_tokens: 32000
       max_output_tokens: 8192
+      pass_max_tokens: true
       input_price: 0.05
       output_price: 0.25
     - name: mistralai/mixtral-8x7b-instruct-v0.1
       max_input_tokens: 32000
       max_output_tokens: 8192
+      pass_max_tokens: true
       input_price: 0.3
       output_price: 1
 
@@ -382,26 +407,31 @@
     - name: ernie-4.0-8k-preview
       max_input_tokens: 5120
       max_output_tokens: 2048
+      pass_max_tokens: true
       input_price: 16.8
       output_price: 16.8
     - name: ernie-3.5-8k-preview
       max_input_tokens: 5120
       max_output_tokens: 2048
+      pass_max_tokens: true
       input_price: 1.68
       output_price: 1.68
     - name: ernie-speed-128k
       max_input_tokens: 124000
       max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 0.56
       output_price: 1.12
     - name: ernie-lite-8k
       max_input_tokens: 7168
       max_output_tokens: 2048
+      pass_max_tokens: true
       input_price: 0.42
       output_price: 0.84
     - name: ernie-tiny-8k
       max_input_tokens: 7168
       max_output_tokens: 2048
+      pass_max_tokens: true
       input_price: 0.14
       output_price: 0.14
 
@@ -414,22 +444,22 @@
   models:
     - name: qwen-turbo
       max_input_tokens: 6000
-      max_output_tokens?: 1500
+      max_output_tokens: 1500
       input_price: 1.12
       output_price: 1.12
     - name: qwen-plus
       max_input_tokens: 30000
-      max_output_tokens?: 2000
+      max_output_tokens: 2000
       input_price: 2.8
       output_price: 2.8
     - name: qwen-max
       max_input_tokens: 6000
-      max_output_tokens?: 2000
+      max_output_tokens: 2000
       input_price: 16.8
       output_price: 16.8
     - name: qwen-max-longcontext
       max_input_tokens: 28000
-      max_output_tokens?: 2000
+      max_output_tokens: 2000
     - name: qwen-vl-plus
       input_price: 1.12
       output_price: 1.12
@@ -686,16 +716,22 @@
       supports_vision: true
     - name: anthropic/claude-3-opus
       max_input_tokens: 200000
+      max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 15
       output_price: 75
       supports_vision: true
     - name: anthropic/claude-3-sonnet
       max_input_tokens: 200000
+      max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 3
       output_price: 15
       supports_vision: true
     - name: anthropic/claude-3-haiku
       max_input_tokens: 200000
+      max_output_tokens: 4096
+      pass_max_tokens: true
       input_price: 0.25
       output_price: 1.25
       supports_vision: true

src/client/bedrock.rs

@@ -172,7 +172,7 @@ async fn send_message_streaming(
                     let data: Value = decode_chunk(message.payload()).ok_or_else(|| {
                         anyhow!("Invalid chunk data: {}", hex_encode(message.payload()))
                     })?;
-                    debug!("bedrock chunk: {data}");
+                    // debug!("bedrock chunk: {data}");
                     match model_category {
                         ModelCategory::Anthropic => {
                             if let Some(typ) = data["type"].as_str() {
@@ -235,7 +235,7 @@ fn meta_llama_build_body(data: SendData, model: &Model, pt: PromptFormat) -> Res
     let prompt = generate_prompt(&messages, pt)?;
     let mut body = json!({ "prompt": prompt });
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["max_gen_len"] = v.into();
     }
     if let Some(v) = temperature {
@@ -258,7 +258,7 @@ fn mistral_build_body(data: SendData, model: &Model) -> Result<Value> {
     let prompt = generate_prompt(&messages, MISTRAL_PROMPT_FORMAT)?;
     let mut body = json!({ "prompt": prompt });
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["max_tokens"] = v.into();
     }
     if let Some(v) = temperature {

src/client/claude.rs

@@ -142,7 +142,7 @@ pub fn claude_build_body(data: SendData, model: &Model) -> Result<Value> {
     if let Some(v) = system_message {
         body["system"] = v.into();
     }
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["max_tokens"] = v.into();
     }
     if let Some(v) = temperature {

src/client/cloudflare.rs

@@ -88,7 +88,7 @@ fn build_body(data: SendData, model: &Model) -> Result<Value> {
         "messages": messages,
     });
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["max_tokens"] = v.into();
     }
     if let Some(v) = temperature {

src/client/cohere.rs

@@ -135,7 +135,7 @@ fn build_body(data: SendData, model: &Model) -> Result<Value> {
         body["chat_history"] = messages.into();
     }
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["max_tokens"] = v.into();
     }
     if let Some(v) = temperature {

src/client/ernie.rs

@@ -128,7 +128,7 @@ fn build_body(data: SendData, model: &Model) -> Value {
         "messages": messages,
     });
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["max_output_tokens"] = v.into();
     }
     if let Some(v) = temperature {

src/client/model.rs

@@ -14,11 +14,11 @@ pub struct Model {
     pub name: String,
     pub max_input_tokens: Option<usize>,
     pub max_output_tokens: Option<isize>,
-    pub ref_max_output_tokens: Option<isize>,
+    pub pass_max_tokens: bool,
     pub input_price: Option<f64>,
     pub output_price: Option<f64>,
-    pub extra_fields: Option<serde_json::Map<String, serde_json::Value>>,
     pub capabilities: ModelCapabilities,
+    pub extra_fields: Option<serde_json::Map<String, serde_json::Value>>,
 }
 
 impl Default for Model {
@@ -32,13 +32,13 @@ impl Model {
         Self {
             client_name: client_name.into(),
             name: name.into(),
-            extra_fields: None,
             max_input_tokens: None,
             max_output_tokens: None,
-            ref_max_output_tokens: None,
+            pass_max_tokens: false,
             input_price: None,
             output_price: None,
             capabilities: ModelCapabilities::Text,
+            extra_fields: None,
         }
     }
 
@@ -49,8 +49,7 @@ impl Model {
                 let mut model = Model::new(client_name, &v.name);
                 model
                     .set_max_input_tokens(v.max_input_tokens)
-                    .set_max_output_tokens(v.max_output_tokens)
-                    .set_ref_max_output_tokens(v.ref_max_output_tokens)
+                    .set_max_tokens(v.max_output_tokens, v.pass_max_tokens)
                     .set_input_price(v.input_price)
                     .set_output_price(v.output_price)
                     .set_supports_vision(v.supports_vision)
@@ -97,7 +96,7 @@ impl Model {
 
     pub fn description(&self) -> String {
         let max_input_tokens = format_option_value(&self.max_input_tokens);
-        let max_output_tokens = format_option_value(&self.show_max_output_tokens());
+        let max_output_tokens = format_option_value(&self.max_output_tokens);
         let input_price = format_option_value(&self.input_price);
         let output_price = format_option_value(&self.output_price);
         let vision = if self.capabilities.contains(ModelCapabilities::Vision) {
@@ -115,8 +114,12 @@ impl Model {
         self.capabilities.contains(ModelCapabilities::Vision)
     }
 
-    pub fn show_max_output_tokens(&self) -> Option<isize> {
-        self.max_output_tokens.or(self.ref_max_output_tokens)
+    pub fn max_tokens_param(&self) -> Option<isize> {
+        if self.pass_max_tokens {
+            self.max_output_tokens
+        } else {
+            None
+        }
     }
 
     pub fn set_max_input_tokens(&mut self, max_input_tokens: Option<usize>) -> &mut Self {
@@ -127,19 +130,16 @@ impl Model {
         self
     }
 
-    pub fn set_max_output_tokens(&mut self, max_output_tokens: Option<isize>) -> &mut Self {
+    pub fn set_max_tokens(
+        &mut self,
+        max_output_tokens: Option<isize>,
+        pass_max_tokens: bool,
+    ) -> &mut Self {
         match max_output_tokens {
             None | Some(0) => self.max_output_tokens = None,
             _ => self.max_output_tokens = max_output_tokens,
         }
-        self
-    }
-
-    pub fn set_ref_max_output_tokens(&mut self, ref_max_output_tokens: Option<isize>) -> &mut Self {
-        match ref_max_output_tokens {
-            None | Some(0) => self.ref_max_output_tokens = None,
-            _ => self.ref_max_output_tokens = ref_max_output_tokens,
-        }
+        self.pass_max_tokens = pass_max_tokens;
         self
     }
 
@@ -237,12 +237,12 @@ pub struct ModelConfig {
     pub name: String,
     pub max_input_tokens: Option<usize>,
     pub max_output_tokens: Option<isize>,
-    #[serde(rename = "max_output_tokens?")]
-    pub ref_max_output_tokens: Option<isize>,
     pub input_price: Option<f64>,
     pub output_price: Option<f64>,
     #[serde(default)]
     pub supports_vision: bool,
+    #[serde(default)]
+    pub pass_max_tokens: bool,
     pub extra_fields: Option<serde_json::Map<String, serde_json::Value>>,
 }
 

src/client/ollama.rs

@@ -159,7 +159,7 @@ fn build_body(data: SendData, model: &Model) -> Result<Value> {
         "options": {},
     });
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["options"]["num_predict"] = v.into();
     }
     if let Some(v) = temperature {

src/client/openai.rs

@@ -90,7 +90,7 @@ pub fn openai_build_body(data: SendData, model: &Model) -> Value {
         "messages": messages,
     });
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["max_tokens"] = v.into();
     }
     if let Some(v) = temperature {

src/client/qianwen.rs

@@ -173,7 +173,7 @@ fn build_body(data: SendData, model: &Model, is_vl: bool) -> Result<(Value, bool
         parameters["incremental_output"] = true.into();
     }
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         parameters["max_tokens"] = v.into();
     }
     if let Some(v) = temperature {

src/client/replicate.rs

@@ -148,7 +148,7 @@ fn build_body(data: SendData, model: &Model) -> Result<Value> {
         "prompt_template": "{prompt}"
     });
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         input["max_tokens"] = v.into();
         input["max_new_tokens"] = v.into();
     }

src/client/vertexai.rs

@@ -201,7 +201,7 @@ pub(crate) fn gemini_build_body(
         body["safetySettings"] = safety_settings;
     }
 
-    if let Some(v) = model.max_output_tokens {
+    if let Some(v) = model.max_tokens_param() {
         body["generationConfig"]["maxOutputTokens"] = v.into();
     }
     if let Some(v) = temperature {

src/config/mod.rs

@@ -422,7 +422,7 @@ impl Config {
             (
                 "max_output_tokens",
                 self.model
-                    .max_output_tokens
+                    .max_tokens_param()
                     .map(|v| format!("{v} (current model)"))
                     .unwrap_or_else(|| "-".into()),
             ),
@@ -523,7 +523,7 @@ impl Config {
             (values, args[0])
         } else if args.len() == 2 {
             let values = match args[0] {
-                "max_output_tokens" => match self.model.show_max_output_tokens() {
+                "max_output_tokens" => match self.model.max_output_tokens {
                     Some(v) => vec![v.to_string()],
                     None => vec![],
                 },
@@ -564,7 +564,7 @@ impl Config {
         match key {
             "max_output_tokens" => {
                 let value = parse_value(value)?;
-                self.model.set_max_output_tokens(value);
+                self.model.set_max_tokens(value, true);
             }
             "temperature" => {
                 let value = parse_value(value)?;

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()?;