add qwen3 and fun-asr-nano

tests/test_fun_asr_nano.rs

@@ -0,0 +1,97 @@
+use std::{pin::pin, time::Instant};
+
+use aha::models::{GenerateModel, fun_asr_nano::generate::FunAsrNanoGenerateModel};
+use aha_openai_dive::v1::resources::chat::ChatCompletionParameters;
+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
+    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);
+    let message = r#"
+    {
+        "model": "fun-asr-nano",
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "audio",
+                        "audio_url": 
+                        {
+                            "url": "https://package-release.coderbox.cn/aiway/test/other/%E5%93%AA%E5%90%92.wav"
+                        }
+                    },           
+                    {
+                        "type": "text", 
+                        "text": "语音转写："
+                    }
+                ]
+            }
+        ]
+    }
+    "#;
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut fun_asr_model = FunAsrNanoGenerateModel::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 = fun_asr_model.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 fun_asr_nano_stream() -> Result<()> {
+    // RUST_BACKTRACE=1 cargo test -F cuda fun_asr_nano_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!("{}/FunAudioLLM/Fun-ASR-Nano-2512/", save_dir);
+    let message = r#"
+    {
+        "model": "fun-asr-nano",
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "audio",
+                        "audio_url": 
+                        {
+                            "url": "https://package-release.coderbox.cn/aiway/test/other/%E5%93%AA%E5%90%92.wav"
+                        }
+                    },           
+                    {
+                        "type": "text", 
+                        "text": "语音转写："
+                    }
+                ]
+            }
+        ]
+    }
+    "#;
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut fun_asr_model = FunAsrNanoGenerateModel::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!(fun_asr_model.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_glm_asr_nano.rs

@@ -70,7 +70,7 @@ async fn glm_asr_nano_stream() -> Result<()> {
                         "type": "audio",
                         "audio_url": 
                         {
-                            "url": "file://./assets/audio/zh.mp3"
+                            "url": "https://package-release.coderbox.cn/aiway/test/other/%E5%93%AA%E5%90%92.wav"
                         }
                     },           
                     {

tests/test_paddleocr_vl.rs

@@ -19,7 +19,7 @@ fn paddleocr_vl_generate() -> Result<()> {
                         "type": "image",
                         "image_url": 
                         {
-                            "url": "file://./assets/img/ocr_test1.png"
+                            "url": "https://www.qqxiuzi.cn/zh/shouxie-shufa/welcome.png"
                         }
                     },              
                     {
@@ -69,7 +69,7 @@ async fn paddleocr_vl_stream() -> Result<()> {
                         "type": "image",
                         "image_url": 
                         {
-                            "url": "file://./assets/img/ocr_test1.png"
+                            "url": "https://www.qqxiuzi.cn/zh/shouxie-shufa/welcome.png"
                         }
                     },              
                     {

tests/test_qwen3.rs

@@ -0,0 +1,83 @@
+use std::{pin::pin, time::Instant};
+
+use aha::models::{GenerateModel, qwen3::generate::Qwen3GenerateModel};
+use aha_openai_dive::v1::resources::chat::ChatCompletionParameters;
+use anyhow::Result;
+use rocket::futures::StreamExt;
+
+#[test]
+fn qwen3_0_6b_generate() -> Result<()> {
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda qwen3_0_6b_generate -r -- --nocapture
+    // test with cuda+flash-attn: RUST_BACKTRACE=1 cargo test -F cuda,flash-attn qwen3_0_6b_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/Qwen3-0.6B/", save_dir);
+    let message = r#"
+    {
+        "model": "qwen3",
+        "messages": [
+            {
+                "role": "user",
+                "content": "你吃饭了没"
+            }
+        ]
+    }
+    "#;
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut model = Qwen3GenerateModel::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 result = model.generate(mes)?;
+    let i_duration = i_start.elapsed();
+    println!("generate: \n {:?}", result);
+    if result.usage.is_some() {
+        let num_token = result.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_0_6b_stream() -> Result<()> {
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda qwen3_0_6b_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/Qwen3-0.6B/", save_dir);
+
+    let message = r#"
+    {
+        "model": "qwen3",
+        "messages": [
+            {
+                "role": "user",
+                "content": "你是谁"
+            }
+        ]
+    }
+    "#;
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut model = Qwen3GenerateModel::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!(model.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

@@ -29,7 +29,7 @@ fn qwen3vl_generate() -> 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？"
                     }
                 ]
             }

tests/test_voxcpm.rs

@@ -27,7 +27,7 @@ fn voxcpm_use_message_generate() -> Result<()> {
                         "type": "audio",
                         "audio_url": 
                         {
-                            "url": "https://sis-sample-audio.obs.cn-north-1.myhuaweicloud.com/16k16bit.wav"
+                            "url": "https://package-release.coderbox.cn/aiway/test/other/%E5%93%AA%E5%90%92.wav"
                         }
                     },              
                     {
@@ -37,7 +37,7 @@ fn voxcpm_use_message_generate() -> Result<()> {
                 ]
             }
         ],
-        "metadata": {"prompt_text": "华为致力于把数字世界带给每个人，每个家庭，每个组织，构建万物互联的智能世界。"}
+        "metadata": {"prompt_text": "天雷滚滚我好怕怕，劈得我浑身掉渣渣。突破天劫我笑哈哈，逆天改命我吹喇叭，滴答滴答滴滴答"}
     }
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;

tests/test_voxcpm1_5.rs

@@ -27,17 +27,17 @@ fn voxcpm1_5_use_message_generate() -> Result<()> {
                         "type": "audio",
                         "audio_url": 
                         {
-                            "url": "https://sis-sample-audio.obs.cn-north-1.myhuaweicloud.com/16k16bit.wav"
+                            "url": "https://package-release.coderbox.cn/aiway/test/other/%E5%93%AA%E5%90%92.wav"
                         }
                     },              
                     {
                         "type": "text", 
-                        "text": "VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly realistic speech."
+                        "text": "老大爷我来啦，红红火火恍恍惚惚"
                     }
                 ]
             }
         ],
-        "metadata": {"prompt_text": "华为致力于把数字世界带给每个人，每个家庭，每个组织，构建万物互联的智能世界。"}
+        "metadata": {"prompt_text": "天雷滚滚我好怕怕，劈得我浑身掉渣渣。突破天劫我笑哈哈，逆天改命我吹喇叭，滴答滴答滴滴答"}
     }
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
@@ -72,7 +72,7 @@ fn voxcpm1_5_generate() -> Result<()> {
     let i_start = Instant::now();
     // let generate = voxcpm_generate.generate_simple("太阳当空照，花儿对我笑，小鸟说早早早".to_string())?;
     let generate = voxcpm_generate.inference(
-        "VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly realistic speech.".to_string(),
+        "老大爷我来啦，红红火火恍恍惚惚".to_string(),
         Some("啥子小师叔，打狗还要看主人，你再要继续，我就是你的对手".to_string()),
         Some("file://./assets/audio/voice_01.wav".to_string()),
         // Some("一定被灰太狼给吃了，我已经为他准备好了花圈了".to_string()),

tests/weight_test.rs

@@ -152,3 +152,47 @@ fn glm_asr_nano_weight() -> Result<()> {
     println!("model_list: {:?}", model_list);
     Ok(())
 }
+
+#[test]
+fn fun_asr_nano_weight() -> Result<()> {
+    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);
+    let model_list = find_type_files(&model_path, "pt")?;
+    println!("model_list: {:?}", model_list);
+    // let dev = get_device(None);
+    let mut dict_to_hashmap = HashMap::new();
+    // let mut dtype = candle_core::DType::F32;
+    for m in model_list {
+        let dict = read_all_with_key(m, Some("state_dict"))?;
+        // dtype = dict[0].1.dtype();
+        for (k, v) in dict {
+            if k.contains("model") {
+                println!("key: {}, tensor shape: {:?}", k, v);
+            }
+            dict_to_hashmap.insert(k, v);
+        }
+    }
+    Ok(())
+}
+
+#[test]
+fn qwen3_weight() -> Result<()> {
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Qwen/Qwen3-0.6B/", save_dir);
+    let model_list = find_type_files(&model_path, "safetensors")?;
+
+    let device = Device::Cpu;
+    for m in &model_list {
+        let weights = safetensors::load(m, &device)?;
+        for (key, tensor) in weights.iter() {
+            // if key.contains(".embed_tokens") {
+            //     println!("=== {} === {:?}", key, tensor.shape());
+            // }
+            println!("=== {} === {:?}", key, tensor.shape());
+        }
+    }
+    println!("model_list: {:?}", model_list);
+    Ok(())
+}