add Qwen3.5 model

tests/messy_test.rs

@@ -5,7 +5,7 @@
 // use std::io::{Read, Seek};
 // use std::{io::Cursor, time::Instant};
 
-use aha::utils::interpolate::interpolate_nearest_2d;
+// use aha::utils::tensor_utils::repeat_interleave;
 // use aha_openai_dive::v1::resources::chat::ChatCompletionParameters;
 use anyhow::Result;
 // use byteorder::{LittleEndian, ReadBytesExt};
@@ -15,12 +15,23 @@ use candle_core::Tensor;
 
 #[test]
 fn messy_test() -> Result<()> {
-    // RUST_BACKTRACE=1 cargo test -F cuda messy_test -r -- --nocapture
+    // RUST_BACKTRACE=1 cargo test -F cuda --test messy_test messy_test -r -- --nocapture
     let device = &candle_core::Device::Cpu;
-    let input = Tensor::arange(0.0f32, 25.0f32, device)?.reshape((1, 1, 5, 5))?;
-    println!("input: {}", input);
-    let x_nearest = interpolate_nearest_2d(&input, (10, 10))?;
-    println!("x_nearest: {}", x_nearest);
+    let t1 = Tensor::randn(0.0, 1.0, (16, 9, 64, 128), device)?;
+    let t2 = Tensor::randn(0.0, 1.0, (16, 9, 128, 64), device)?;
+    let out = t1.matmul(&t2)?;
+    println!("out shape: {:?}", out);
+
+    // let input = Tensor::arange(0.0f32, 25.0f32, device)?.reshape((5, 5))?;
+    // println!("input: {}", input);
+    // // let input = input.unsqueeze(D::Minus1)?;
+    // // let input = input.repeat((1, 1, 2))?;
+    // // let input = input.flatten(D::Minus2, D::Minus1)?;
+    // let output = repeat_interleave(&input, 2, 1)?;
+    // println!("output: {}", output);
+
+    // let x_nearest = interpolate_nearest_2d(&input, (10, 10))?;
+    // println!("x_nearest: {}", x_nearest);
     // let input = Tensor::arange(0.0f32, 25.0f32, device)?.reshape((1, 5, 5))?;
     // println!("input: {}", input);
     // let x_nearest = interpolate_nearest_1d(&input, 10)?;

tests/test_fun_asr_nano.rs

@@ -6,7 +6,7 @@ use anyhow::Result;
 use rocket::futures::StreamExt;
 #[test]
 fn fun_asr_nano_generate() -> Result<()> {
-    // RUST_BACKTRACE=1 cargo test -F cuda fun_asr_nano_generate -r -- --nocapture
+    // RUST_BACKTRACE=1 cargo test -F cuda --test test_fun_asr_nano fun_asr_nano_generate -r -- --nocapture
     let save_dir =
         aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
     let model_path = format!("{}/FunAudioLLM/Fun-ASR-Nano-2512/", save_dir);

tests/test_glm_asr_nano.rs

@@ -7,7 +7,7 @@ use rocket::futures::StreamExt;
 
 #[test]
 fn glm_asr_nano_generate() -> Result<()> {
-    // RUST_BACKTRACE=1 cargo test -F cuda glm_asr_nano_generate -r -- --nocapture
+    // RUST_BACKTRACE=1 cargo test -F cuda --test test_glm_asr_nano glm_asr_nano_generate -r -- --nocapture
     let save_dir =
         aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
     let model_path = format!("{}/ZhipuAI/GLM-ASR-Nano-2512/", save_dir);

tests/test_qwen3_5.rs

@@ -0,0 +1,105 @@
+use std::{pin::pin, time::Instant};
+
+use aha::models::{GenerateModel, qwen3_5::generate::Qwen3_5GenerateModel};
+use aha_openai_dive::v1::resources::chat::ChatCompletionParameters;
+use anyhow::Result;
+use rocket::futures::StreamExt;
+
+#[test]
+fn qwen3_5_generate() -> Result<()> {
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda --test test_qwen3_5 qwen3_5_generate -r -- --nocapture
+
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Qwen/Qwen3.5-0.8B/", save_dir);
+
+    let message = r#"
+    {
+        "model": "qwen3.5",
+        "messages": [
+            {
+                "role": "user",
+                "content": [ 
+                    {
+                        "type": "image",
+                        "image_url": 
+                        {
+                            "url": "file:///home/jhq/Downloads/gougou1.jpg"
+                        }
+                    },             
+                    {
+                        "type": "text", 
+                        "text": "描述这张图片."
+                    }
+                ]
+            }
+        ],
+        "metadata": {"enable_thinking": "true"}
+    }
+    "#;
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut qwen3vl = Qwen3_5GenerateModel::init(&model_path, None, None)?;
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in load model is: {:?}", i_duration);
+
+    let i_start = Instant::now();
+    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 qwen3_5_stream() -> Result<()> {
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda --test test_qwen3_5 qwen3_5_stream -r -- --nocapture
+
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Qwen/Qwen3.5-0.8B/", save_dir);
+
+    let message = r#"
+    {
+        "model": "qwen3.5",
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "image",
+                        "image_url": 
+                        {
+                            "url": "file:///home/jhq/Downloads/gougou1.jpg"
+                        }
+                    },             
+                    {
+                        "type": "text", 
+                        "text": "描述这张图片."
+                    }
+                ]
+            }
+        ]
+    }
+    "#;
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut qwen3_5 = Qwen3_5GenerateModel::init(&model_path, None, None)?;
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in load model is: {:?}", i_duration);
+
+    let i_start = Instant::now();
+    let mut stream = pin!(qwen3_5.generate_stream(mes)?);
+    while let Some(item) = stream.next().await {
+        println!("generate: \n {:?}", item);
+    }
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in generate is: {:?}", i_duration);
+    Ok(())
+}

tests/test_qwen3vl.rs

@@ -59,7 +59,7 @@ fn qwen3vl_thinking_generate() -> Result<()> {
 
 #[test]
 fn qwen3vl_generate() -> Result<()> {
-    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda,ffmpeg qwen3vl_generate -r -- --nocapture
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda --test test_qwen3vl qwen3vl_generate -r -- --nocapture
 
     let save_dir =
         aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
@@ -73,15 +73,15 @@ fn qwen3vl_generate() -> Result<()> {
                 "role": "user",
                 "content": [
                     {
-                        "type": "video",
-                        "video_url": 
+                        "type": "image",
+                        "image_url": 
                         {
-                            "url": "./assets/video/video_test.mp4"
+                            "url": "file:///home/jhq/Downloads/gougou1.jpg"
                         }
-                    },              
+                    },             
                     {
                         "type": "text", 
-                        "text": "视频中发生了什么？"
+                        "text": "描述这张图片."
                     }
                 ]
             }