add Qwen3VL model

src/models/qwen2_5vl/generate.rs

@@ -63,7 +63,7 @@ impl<'a> Qwen2_5VLGenerateModel<'a> {
 
 impl<'a> GenerateModel for Qwen2_5VLGenerateModel<'a> {
     fn generate(&mut self, mes: ChatCompletionParameters) -> Result<ChatCompletionResponse> {
-        let mut logit_processor = get_logit_processor(mes.temperature, mes.top_p);
+        let mut logit_processor = get_logit_processor(mes.temperature, mes.top_p, None);
         let mes_render = self.chat_template.apply_chat_template(&mes)?;
         let input = self.pre_processor.process_info(&mes, &mes_render)?;
         let mut input_ids = self
@@ -118,11 +118,12 @@ impl<'a> GenerateModel for Qwen2_5VLGenerateModel<'a> {
         let response = build_completion_response(res, "qwen2.5vl");
         Ok(response)
     }
+    
     fn generate_stream(
         &mut self,
         mes: ChatCompletionParameters,
     ) -> Result<impl Stream<Item = Result<ChatCompletionChunkResponse, anyhow::Error>>> {
-        let mut logit_processor = get_logit_processor(mes.temperature, mes.top_p);
+        let mut logit_processor = get_logit_processor(mes.temperature, mes.top_p, None);
         let mes_render = self.chat_template.apply_chat_template(&mes)?;
         let input = self.pre_processor.process_info(&mes, &mes_render)?;
         let mut input_ids = self

src/models/qwen2_5vl/model.rs

@@ -841,7 +841,7 @@ impl Qwen2_5VLTextModel {
             if seq_len <= 1 {
                 None
             } else {
-                Some(&self.prepare_causal_attention_mask(b_size, seq_len, seqlen_offset)?)
+                Some(&self.prepare_causal_attention_mask(b_size, seq_len, 0)?)
             }
         };
         for layer in self.layers.iter_mut() {