merge pr/KingBright/13

tests/test_deepseek_ocr.rs

@@ -32,10 +32,12 @@ fn deepseek_ocr_generate() -> Result<()> {
         "metadata": {"base_size": "640", "image_size": "640", "crop_mode": "false"}
     }
     "#;
-    let model_path = "/home/jhq/huggingface_model/deepseek-ai/DeepSeek-OCR/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/deepseek-ai/DeepSeek-OCR/", save_dir);
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = DeepseekOCRGenerateModel::init(model_path, None, None)?;
+    let mut model = DeepseekOCRGenerateModel::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();
@@ -84,10 +86,12 @@ async fn deepseek_ocr_stream() -> Result<()> {
         "metadata": {"base_size": "640", "image_size": "640", "crop_mode": "false"}
     }
     "#;
-    let model_path = "/home/jhq/huggingface_model/deepseek-ai/DeepSeek-OCR/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/deepseek-ai/DeepSeek-OCR/", save_dir);
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = DeepseekOCRGenerateModel::init(model_path, None, None)?;
+    let mut model = DeepseekOCRGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
     let mut stream = pin!(model.generate_stream(mes)?);

tests/test_gelab_zero.rs

@@ -8,7 +8,9 @@ use anyhow::Result;
 fn gelab_zero_generate() -> Result<()> {
     // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda gelab_zero_generate -r -- --nocapture
 
-    let model_path = "/home/jhq/huggingface_model/stepfun-ai/GELab-Zero-4B-preview";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/stepfun-ai/GELab-Zero-4B-preview", save_dir);
 
     let message = r#"
     {
@@ -56,7 +58,7 @@ fn gelab_zero_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut qwen3vl = Qwen3VLGenerateModel::init(model_path, None, None)?;
+    let mut qwen3vl = Qwen3VLGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 

tests/test_glm_asr_nano.rs

@@ -8,7 +8,9 @@ use rocket::futures::StreamExt;
 #[test]
 fn glm_asr_nano_generate() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda glm_asr_nano_generate -r -- --nocapture
-    let model_path = "/home/jhq/huggingface_model/ZhipuAI/GLM-ASR-Nano-2512/";
+    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);
     let message = r#"
     {
         "model": "glm-asr-nano",
@@ -34,7 +36,7 @@ fn glm_asr_nano_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut glm_asr_model = GlmAsrNanoGenerateModel::init(model_path, None, None)?;
+    let mut glm_asr_model = GlmAsrNanoGenerateModel::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();
@@ -54,7 +56,9 @@ fn glm_asr_nano_generate() -> Result<()> {
 #[tokio::test]
 async fn glm_asr_nano_stream() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda glm_asr_nano_stream -r -- --nocapture
-    let model_path = "/home/jhq/huggingface_model/ZhipuAI/GLM-ASR-Nano-2512/";
+    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);
     let message = r#"
     {
         "model": "glm-asr-nano",
@@ -80,7 +84,7 @@ async fn glm_asr_nano_stream() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut glm_asr_model = GlmAsrNanoGenerateModel::init(model_path, None, None)?;
+    let mut glm_asr_model = GlmAsrNanoGenerateModel::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();

tests/test_hunyuan_ocr.rs

@@ -31,10 +31,12 @@ fn hunyuan_ocr_generate() -> Result<()> {
         ]
     }
     "#;
-    let model_path = "/home/jhq/huggingface_model/Tencent-Hunyuan/HunyuanOCR/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Tencent-Hunyuan/HunyuanOCR/", save_dir);
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = HunyuanOCRGenerateModel::init(model_path, None, None)?;
+    let mut model = HunyuanOCRGenerateModel::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();
@@ -79,10 +81,12 @@ async fn hunyuan_ocr_stream() -> Result<()> {
         ]
     }
     "#;
-    let model_path = "/home/jhq/huggingface_model/Tencent-Hunyuan/HunyuanOCR/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Tencent-Hunyuan/HunyuanOCR/", save_dir);
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = HunyuanOCRGenerateModel::init(model_path, None, None)?;
+    let mut model = HunyuanOCRGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
     let mut stream = pin!(model.generate_stream(mes)?);

tests/test_minicpm4.rs

@@ -11,7 +11,9 @@ fn minicpm_generate() -> Result<()> {
     // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda minicpm_generate -r -- --nocapture
     // test with cuda+flash-attn: RUST_BACKTRACE=1 cargo test -F cuda,flash-attn minicpm_generate -r -- --nocapture
 
-    let model_path = "/home/jhq/huggingface_model/OpenBMB/MiniCPM4-0.5B/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/MiniCPM4-0.5B/", save_dir);
     let message = r#"
     {
         "temperature": 0.3,
@@ -27,7 +29,7 @@ fn minicpm_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = MiniCPMGenerateModel::init(model_path, None, None)?;
+    let mut model = MiniCPMGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 
@@ -50,7 +52,9 @@ fn minicpm_generate() -> Result<()> {
 async fn minicpm_stream() -> Result<()> {
     // test with cuda+flash-attn: RUST_BACKTRACE=1 cargo test -F cuda,flash-attn minicpm_stream -r -- --nocapture
 
-    let model_path = "/home/jhq/huggingface_model/OpenBMB/MiniCPM4-0.5B/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/MiniCPM4-0.5B/", save_dir);
 
     let message = r#"
     {
@@ -65,7 +69,7 @@ async fn minicpm_stream() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = MiniCPMGenerateModel::init(model_path, None, None)?;
+    let mut model = MiniCPMGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 

tests/test_paddleocr_vl.rs

@@ -32,10 +32,12 @@ fn paddleocr_vl_generate() -> Result<()> {
         "stream": false
     }
     "#;
-    let model_path = "/home/jhq/huggingface_model/PaddlePaddle/PaddleOCR-VL/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/PaddlePaddle/PaddleOCR-VL/", save_dir);
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = PaddleOCRVLGenerateModel::init(model_path, None, None)?;
+    let mut model = PaddleOCRVLGenerateModel::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();
@@ -79,10 +81,12 @@ async fn paddleocr_vl_stream() -> Result<()> {
         ]
     }
     "#;
-    let model_path = "/home/jhq/huggingface_model/PaddlePaddle/PaddleOCR-VL/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/PaddlePaddle/PaddleOCR-VL/", save_dir);
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = PaddleOCRVLGenerateModel::init(model_path, None, None)?;
+    let mut model = PaddleOCRVLGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
     let mut stream = pin!(model.generate_stream(mes)?);

tests/test_qwen2_5vl.rs

@@ -13,7 +13,9 @@ fn qwen2_5vl_generate() -> Result<()> {
     // let device = Device::cuda_if_available(0)?;
     // let dtype = DType::BF16;
 
-    let model_path = "/home/jhq/huggingface_model/Qwen/Qwen2.5-VL-3B-Instruct/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Qwen/Qwen2.5-VL-3B-Instruct/", save_dir);
 
     let message = r#"
     {
@@ -40,7 +42,7 @@ fn qwen2_5vl_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = Qwen2_5VLGenerateModel::init(model_path, None, None)?;
+    let mut model = Qwen2_5VLGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 
@@ -65,7 +67,9 @@ async fn qwen2_5vl_stream() -> Result<()> {
     // let device = Device::cuda_if_available(0)?;
     // let dtype = DType::BF16;
 
-    let model_path = "/home/jhq/huggingface_model/Qwen/Qwen2.5-VL-3B-Instruct/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Qwen/Qwen2.5-VL-3B-Instruct/", save_dir);
 
     let message = r#"
     {
@@ -92,7 +96,7 @@ async fn qwen2_5vl_stream() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = Qwen2_5VLGenerateModel::init(model_path, None, None)?;
+    let mut model = Qwen2_5VLGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 

tests/test_qwen3vl.rs

@@ -9,7 +9,9 @@ use rocket::futures::StreamExt;
 fn qwen3vl_generate() -> Result<()> {
     // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda,ffmpeg qwen3vl_generate -r -- --nocapture
 
-    let model_path = "/home/jhq/huggingface_model/Qwen/Qwen3-VL-2B-Instruct/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Qwen/Qwen3-VL-2B-Instruct/", save_dir);
 
     let message = r#"
     {
@@ -36,7 +38,7 @@ fn qwen3vl_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut qwen3vl = Qwen3VLGenerateModel::init(model_path, None, None)?;
+    let mut qwen3vl = Qwen3VLGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 
@@ -58,7 +60,9 @@ fn qwen3vl_generate() -> Result<()> {
 async fn qwen3vl_stream() -> Result<()> {
     // 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/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Qwen/Qwen3-VL-2B-Instruct/", save_dir);
 
     let message = r#"
     {
@@ -85,7 +89,7 @@ async fn qwen3vl_stream() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut qwen3vl = Qwen3VLGenerateModel::init(model_path, None, None)?;
+    let mut qwen3vl = Qwen3VLGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 

tests/test_rmbg2_0.rs

@@ -8,7 +8,9 @@ use anyhow::Result;
 fn rmbg2_0_generate() -> Result<()> {
     // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda rmbg2_0_generate -r -- --nocapture
 
-    let model_path = "/home/jhq/huggingface_model/AI-ModelScope/RMBG-2.0/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/AI-ModelScope/RMBG-2.0/", save_dir);
 
     let message = r#"
     {
@@ -31,7 +33,7 @@ fn rmbg2_0_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let model = RMBG2_0Model::init(model_path, None, None)?;
+    let model = RMBG2_0Model::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 

tests/test_robo_brain.rs

@@ -8,7 +8,9 @@ use anyhow::Result;
 fn robo_brain_generate() -> Result<()> {
     // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda robo_brain_generate -r -- --nocapture
 
-    let model_path = "/home/jhq/huggingface_model/BAAI/RoboBrain2.0-3B/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/BAAI/RoboBrain2.0-3B/", save_dir);
 
     let message = r#"
     {
@@ -28,7 +30,7 @@ fn robo_brain_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut model = Qwen2_5VLGenerateModel::init(model_path, None, None)?;
+    let mut model = Qwen2_5VLGenerateModel::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 

tests/test_voxcpm.rs

@@ -13,7 +13,9 @@ use anyhow::{Ok, Result};
 #[test]
 fn voxcpm_use_message_generate() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda voxcpm_use_message_generate -r -- --nocapture
-    let model_path = "/home/jhq/huggingface_model/openbmb/VoxCPM-0.5B/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/VoxCPM-0.5B/", save_dir);
     let message = r#"
     {
         "model": "voxcpm",
@@ -40,7 +42,7 @@ fn voxcpm_use_message_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut voxcpm_generate = VoxCPMGenerate::init(model_path, None, None)?;
+    let mut voxcpm_generate = VoxCPMGenerate::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 
@@ -59,10 +61,12 @@ fn voxcpm_use_message_generate() -> Result<()> {
 #[test]
 fn voxcpm_generate() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda voxcpm_generate -r -- --nocapture
-    let model_path = "/home/jhq/huggingface_model/openbmb/VoxCPM-0.5B/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/VoxCPM-0.5B/", save_dir);
 
     let i_start = Instant::now();
-    let mut voxcpm_generate = VoxCPMGenerate::init(model_path, None, None)?;
+    let mut voxcpm_generate = VoxCPMGenerate::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 
@@ -106,8 +110,10 @@ fn voxcpm_generate() -> Result<()> {
 
 #[test]
 fn voxcpm_tokenizer() -> Result<()> {
-    let model_path = "/home/jhq/huggingface_model/openbmb/VoxCPM-0.5B/";
-    let tokenizer = SingleChineseTokenizer::new(model_path)?;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/VoxCPM-0.5B/", save_dir);
+    let tokenizer = SingleChineseTokenizer::new(&model_path)?;
     let ids = tokenizer.encode("你好啊，你吃饭了吗".to_string())?;
     println!("ids: {:?}", ids);
     Ok(())

tests/test_voxcpm1_5.rs

@@ -13,7 +13,9 @@ use anyhow::{Ok, Result};
 #[test]
 fn voxcpm1_5_use_message_generate() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda voxcpm1_5_use_message_generate -r -- --nocapture
-    let model_path = "/home/jhq/huggingface_model/OpenBMB/VoxCPM1.5/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/VoxCPM1.5/", save_dir);
     let message = r#"
     {
         "model": "voxcpm1.5",
@@ -40,7 +42,7 @@ fn voxcpm1_5_use_message_generate() -> Result<()> {
     "#;
     let mes: ChatCompletionParameters = serde_json::from_str(message)?;
     let i_start = Instant::now();
-    let mut voxcpm_generate = VoxCPMGenerate::init(model_path, None, None)?;
+    let mut voxcpm_generate = VoxCPMGenerate::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 
@@ -58,10 +60,12 @@ fn voxcpm1_5_use_message_generate() -> Result<()> {
 #[test]
 fn voxcpm1_5_generate() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda voxcpm1_5_generate -r -- --nocapture
-    let model_path = "/home/jhq/huggingface_model/OpenBMB/VoxCPM1.5/";
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/VoxCPM1.5/", save_dir);
 
     let i_start = Instant::now();
-    let mut voxcpm_generate = VoxCPMGenerate::init(model_path, None, None)?;
+    let mut voxcpm_generate = VoxCPMGenerate::init(&model_path, None, None)?;
     let i_duration = i_start.elapsed();
     println!("Time elapsed in load model is: {:?}", i_duration);
 
@@ -106,8 +110,10 @@ fn voxcpm1_5_generate() -> Result<()> {
 #[test]
 fn voxcpm1_5_tokenizer() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda voxcpm1_5_tokenizer -r -- --nocapture
-    let model_path = "/home/jhq/huggingface_model/OpenBMB/VoxCPM1.5/";
-    let tokenizer = SingleChineseTokenizer::new(model_path)?;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/VoxCPM1.5/", save_dir);
+    let tokenizer = SingleChineseTokenizer::new(&model_path)?;
     let ids = tokenizer.encode("你好啊，你吃饭了吗".to_string())?;
     println!("ids: {:?}", ids);
     Ok(())

tests/weight_test.rs

@@ -7,8 +7,10 @@ use candle_nn::VarBuilder;
 
 #[test]
 fn minicpm4_weight() -> Result<()> {
-    let model_path = "/home/jhq/huggingface_model/OpenBMB/MiniCPM4-0.5B/";
-    let model_list = find_type_files(model_path, "safetensors")?;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/MiniCPM4-0.5B/", 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)?;
@@ -23,8 +25,10 @@ fn minicpm4_weight() -> Result<()> {
 
 #[test]
 fn voxcpm_weight() -> Result<()> {
-    let model_path = "/home/jhq/huggingface_model/openbmb/VoxCPM-0.5B/";
-    let model_list = find_type_files(model_path, "pth")?;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/VoxCPM-0.5B/", save_dir);
+    let model_list = find_type_files(&model_path, "pth")?;
     println!("model_list: {:?}", model_list);
     let dev = get_device(None);
     let mut dict_to_hashmap = HashMap::new();
@@ -48,8 +52,10 @@ fn voxcpm_weight() -> Result<()> {
 
 #[test]
 fn voxcpm1_5_weight() -> Result<()> {
-    let model_path = "/home/jhq/huggingface_model/OpenBMB/VoxCPM1.5/";
-    let model_list = find_type_files(model_path, "pth")?;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/VoxCPM1.5/", save_dir);
+    let model_list = find_type_files(&model_path, "pth")?;
     println!("model_list: {:?}", model_list);
     // let dev = get_device(None);
     let mut dict_to_hashmap = HashMap::new();
@@ -68,8 +74,10 @@ fn voxcpm1_5_weight() -> Result<()> {
 
 #[test]
 fn qwen3vl_weight() -> Result<()> {
-    let model_path = "/home/jhq/huggingface_model/Qwen/Qwen3-VL-4B-Instruct/";
-    let model_list = find_type_files(model_path, "safetensors")?;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Qwen/Qwen3-VL-4B-Instruct/", save_dir);
+    let model_list = find_type_files(&model_path, "safetensors")?;
 
     let device = Device::Cpu;
     for m in &model_list {
@@ -84,8 +92,10 @@ fn qwen3vl_weight() -> Result<()> {
 
 #[test]
 fn deepseekocr_weight() -> Result<()> {
-    let model_path = "/home/jhq/huggingface_model/deepseek-ai/DeepSeek-OCR/";
-    let model_list = find_type_files(model_path, "safetensors")?;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/deepseek-ai/DeepSeek-OCR/", save_dir);
+    let model_list = find_type_files(&model_path, "safetensors")?;
 
     let device = Device::Cpu;
     for m in &model_list {
@@ -103,8 +113,10 @@ fn deepseekocr_weight() -> Result<()> {
 
 #[test]
 fn hunyuanocr_weight() -> Result<()> {
-    let model_path = "/home/jhq/huggingface_model/Tencent-Hunyuan/HunyuanOCR/";
-    let model_list = find_type_files(model_path, "safetensors")?;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/Tencent-Hunyuan/HunyuanOCR/", save_dir);
+    let model_list = find_type_files(&model_path, "safetensors")?;
 
     let device = Device::Cpu;
     for m in &model_list {