model test add tps

tests/test_qwen3vl.rs

@@ -19,15 +19,15 @@ fn qwen3vl_generate() -> Result<()> {
                 "role": "user",
                 "content": [
                     {
-                        "type": "image",
-                        "image_url": 
+                        "type": "video",
+                        "video_url": 
                         {
-                            "url": "file://./assets/img/ocr_test1.png"
+                            "url": "./assets/video/video_test.mp4"
                         }
-                    },             
+                    },              
                     {
                         "type": "text", 
-                        "text": "请分析图片并提取所有可见文本内容，按从左到右、从上到下的布局，返回纯文本"
+                        "text": "视频中发生了什么？, 现在几点了"
                     }
                 ]
             }
@@ -44,13 +44,19 @@ fn qwen3vl_generate() -> Result<()> {
     let res = qwen3vl.generate(mes)?;
     let i_duration = i_start.elapsed();
     println!("generate: \n {:?}", res);
+    if res.usage.is_some() {
+        let num_token = res.usage.as_ref().unwrap().total_tokens;
+        let duration_secs = i_duration.as_secs_f64();
+        let tps = num_token as f64 / duration_secs;
+        println!("Tokens per second (TPS): {:.2}", tps);
+    }
     println!("Time elapsed in generate is: {:?}", i_duration);
     Ok(())
 }
 
 #[tokio::test]
 async fn qwen3vl_stream() -> Result<()> {
-    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda qwen3vl_stream -r -- --nocapture
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda,ffmpeg qwen3vl_stream -r -- --nocapture
 
     let model_path = "/home/jhq/huggingface_model/Qwen/Qwen3-VL-2B-Instruct/";
 
@@ -60,7 +66,7 @@ async fn qwen3vl_stream() -> Result<()> {
         "messages": [
             {
                 "role": "user",
-                "content": [ 
+                "content": [
                     {
                         "type": "video",
                         "video_url": 
@@ -70,39 +76,11 @@ async fn qwen3vl_stream() -> Result<()> {
                     },              
                     {
                         "type": "text", 
-                        "text": "视频中发生了什么？, 现在几点了"
+                        "text": "视频中发生了什么？"
                     }
                 ]
             }
-        ],
-        "tools": [
-            {
-                "type": "function",
-                "function": {
-                    "name": "get_current_time",
-                    "description": "当你想知道现在的时间时非常有用。",
-                    "parameters": {}
-                }
-            },
-            {
-                "type": "function",
-                "function": {
-                    "name": "get_current_weather",
-                    "description": "当你想查询指定城市的天气时非常有用。",
-                    "parameters": {
-                        "type": "object",
-                        "properties": {
-                            "location": {
-                                "type": "string",
-                                "description": "城市或县区，比如北京市、杭州市、余杭区等。"
-                            }
-                        },
-                        "required": ["location"]
-                    }
-                }
-            }
-        ],
-        "tool_choice": null
+        ]
     }
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;