dynamic home dir & model donwload script

Cargo.lock

@@ -377,6 +377,12 @@ version = "2.6.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6099cdc01846bc367c4e7dd630dc5966dccf36b652fae7a74e17b640411a91b2"
 
+[[package]]
+name = "block"
+version = "0.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0d8c1fef690941d3e7788d328517591fecc684c084084702d6ff1641e993699a"
+
 [[package]]
 name = "built"
 version = "0.7.7"
@@ -438,10 +444,12 @@ checksum = "a9f51e2ecf6efe9737af8f993433c839f956d2b6ed4fd2dd4a7c6d8b0fa667ff"
 dependencies = [
  "byteorder",
  "candle-kernels",
+ "candle-metal-kernels",
  "cudarc",
  "gemm 0.17.1",
  "half",
  "memmap2",
+ "metal 0.27.0",
  "num-traits",
  "num_cpus",
  "rand 0.9.2",
@@ -451,6 +459,7 @@ dependencies = [
  "thiserror 1.0.69",
  "ug",
  "ug-cuda",
+ "ug-metal",
  "yoke 0.7.5",
  "zip",
 ]
@@ -476,6 +485,19 @@ dependencies = [
  "bindgen_cuda",
 ]
 
+[[package]]
+name = "candle-metal-kernels"
+version = "0.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9a323ee9c813707f73b6e59300661b354a70410f31fe4135170c4eda8a061534"
+dependencies = [
+ "half",
+ "metal 0.27.0",
+ "once_cell",
+ "thiserror 1.0.69",
+ "tracing",
+]
+
 [[package]]
 name = "candle-nn"
 version = "0.9.1"
@@ -483,7 +505,9 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c1980d53280c8f9e2c6cbe1785855d7ff8010208b46e21252b978badf13ad69d"
 dependencies = [
  "candle-core",
+ "candle-metal-kernels",
  "half",
+ "metal 0.27.0",
  "num-traits",
  "rayon",
  "safetensors",
@@ -700,6 +724,17 @@ version = "0.8.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "773648b94d0e5d620f64f280777445740e61fe701025087ec8b57f45c791888b"
 
+[[package]]
+name = "core-graphics-types"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "45390e6114f68f718cc7a830514a96f903cccd70d02a8f6d9f643ac4ba45afaf"
+dependencies = [
+ "bitflags 1.3.2",
+ "core-foundation",
+ "libc",
+]
+
 [[package]]
 name = "crc32fast"
 version = "1.5.0"
@@ -1153,7 +1188,28 @@ version = "0.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f6f339eb8adc052cd2ca78910fda869aefa38d22d5cb648e6485e4d3fc06f3b1"
 dependencies = [
- "foreign-types-shared",
+ "foreign-types-shared 0.1.1",
+]
+
+[[package]]
+name = "foreign-types"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d737d9aa519fb7b749cbc3b962edcf310a8dd1f4b67c91c4f83975dbdd17d965"
+dependencies = [
+ "foreign-types-macros",
+ "foreign-types-shared 0.3.1",
+]
+
+[[package]]
+name = "foreign-types-macros"
+version = "0.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1a5c6c585bc94aaf2c7b51dd4c2ba22680844aba4c687be581871a6f518c5742"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
 ]
 
 [[package]]
@@ -1162,6 +1218,12 @@ version = "0.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "00b0228411908ca8685dba7fc2cdd70ec9990a6e753e89b6ac91a84c40fbaf4b"
 
+[[package]]
+name = "foreign-types-shared"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "aa9a19cbb55df58761df49b23516a86d432839add4af60fc256da840f66ed35b"
+
 [[package]]
 name = "form_urlencoded"
 version = "1.2.2"
@@ -2249,6 +2311,15 @@ version = "0.2.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "670fdfda89751bc4a84ac13eaa63e205cf0fd22b4c9a5fbfa085b63c1f1d3a30"
 
+[[package]]
+name = "malloc_buf"
+version = "0.0.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "62bb907fe88d54d8d9ce32a3cceab4218ed2f6b7d35617cafe9adf84e43919cb"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "matchers"
 version = "0.2.0"
@@ -2284,6 +2355,36 @@ dependencies = [
  "stable_deref_trait",
 ]
 
+[[package]]
+name = "metal"
+version = "0.27.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c43f73953f8cbe511f021b58f18c3ce1c3d1ae13fe953293e13345bf83217f25"
+dependencies = [
+ "bitflags 2.10.0",
+ "block",
+ "core-graphics-types",
+ "foreign-types 0.5.0",
+ "log",
+ "objc",
+ "paste",
+]
+
+[[package]]
+name = "metal"
+version = "0.29.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7ecfd3296f8c56b7c1f6fbac3c71cefa9d78ce009850c45000015f206dc7fa21"
+dependencies = [
+ "bitflags 2.10.0",
+ "block",
+ "core-graphics-types",
+ "foreign-types 0.5.0",
+ "log",
+ "objc",
+ "paste",
+]
+
 [[package]]
 name = "mime"
 version = "0.3.17"
@@ -2590,6 +2691,25 @@ version = "0.4.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "830b246a0e5f20af87141b25c173cd1b609bd7779a4617d6ec582abaf90870f3"
 
+[[package]]
+name = "objc"
+version = "0.2.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "915b1b472bc21c53464d6c8461c9d3af805ba1ef837e1cac254428f4a77177b1"
+dependencies = [
+ "malloc_buf",
+ "objc_exception",
+]
+
+[[package]]
+name = "objc_exception"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ad970fb455818ad6cba4c122ad012fae53ae8b4795f86378bce65e4f6bab2ca4"
+dependencies = [
+ "cc",
+]
+
 [[package]]
 name = "object"
 version = "0.37.3"
@@ -2641,7 +2761,7 @@ checksum = "24ad14dd45412269e1a30f52ad8f0664f0f4f4a89ee8fe28c3b3527021ebb654"
 dependencies = [
  "bitflags 2.10.0",
  "cfg-if",
- "foreign-types",
+ "foreign-types 0.3.2",
  "libc",
  "once_cell",
  "openssl-macros",
@@ -4236,6 +4356,20 @@ dependencies = [
  "ug",
 ]
 
+[[package]]
+name = "ug-metal"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "76daec3c7a32a1b4a0e3307b6b057fa067aa64e750713987410a2c402e5cd731"
+dependencies = [
+ "half",
+ "metal 0.29.0",
+ "objc",
+ "serde",
+ "thiserror 1.0.69",
+ "ug",
+]
+
 [[package]]
 name = "uncased"
 version = "0.9.10"

download_and_run.sh

@@ -0,0 +1,156 @@
+#!/bin/bash
+set -e
+
+# Default values
+# MIRROR_URL="https://hf-mirror.com"
+DEFAULT_SAVE_DIR="$HOME/.aha"
+MODEL_ALIAS=""
+
+# Help function
+show_help() {
+    echo "Usage: $0 [model_alias]"
+    echo ""
+    echo "Arguments:"
+    echo "  model_alias       The model alias to download (e.g., voxcpm, qwen2.5vl-3b)"
+    echo ""
+    echo "Available models:"
+    echo "  minicpm4-0.5b"
+    echo "  qwen2.5vl-3b"
+    echo "  qwen2.5vl-7b"
+    echo "  qwen3vl-2b"
+    echo "  qwen3vl-4b"
+    echo "  qwen3vl-8b"
+    echo "  qwen3vl-32b"
+    echo "  deepseek-ocr"
+    echo "  hunyuan-ocr"
+    echo "  paddleocr-vl"
+    echo "  RMBG2.0"
+    echo "  voxcpm"
+    echo "  voxcpm1.5"
+    echo ""
+    exit 1
+}
+
+# Check if model alias is provided
+if [ -z "$1" ]; then
+    show_help
+fi
+
+MODEL_ALIAS=$1
+
+# Map alias to Repo ID
+MODEL_ID=""
+case $MODEL_ALIAS in
+    "minicpm4-0.5b")
+        MODEL_ID="OpenBMB/MiniCPM4-0.5B"
+        ;;
+    "qwen2.5vl-3b")
+        MODEL_ID="Qwen/Qwen2.5-VL-3B-Instruct"
+        ;;
+    "qwen2.5vl-7b")
+        MODEL_ID="Qwen/Qwen2.5-VL-7B-Instruct"
+        ;;
+    "qwen3vl-2b")
+        MODEL_ID="Qwen/Qwen3-VL-2B-Instruct"
+        ;;
+    "qwen3vl-4b")
+        MODEL_ID="Qwen/Qwen3-VL-4B-Instruct"
+        ;;
+    "qwen3vl-8b")
+        MODEL_ID="Qwen/Qwen3-VL-8B-Instruct"
+        ;;
+    "qwen3vl-32b")
+        MODEL_ID="Qwen/Qwen3-VL-32B-Instruct"
+        ;;
+    "deepseek-ocr")
+        MODEL_ID="deepseek-ai/DeepSeek-OCR"
+        ;;
+    "hunyuan-ocr")
+        MODEL_ID="Tencent-Hunyuan/HunyuanOCR"
+        ;;
+    "paddleocr-vl")
+        MODEL_ID="PaddlePaddle/PaddleOCR-VL"
+        ;;
+    "RMBG2.0")
+        MODEL_ID="AI-ModelScope/RMBG-2.0"
+        ;;
+    "voxcpm")
+        MODEL_ID="OpenBMB/VoxCPM-0.5B"
+        ;;
+    "voxcpm1.5")
+        MODEL_ID="OpenBMB/VoxCPM1.5"
+        ;;
+    *)
+        echo "Error: Unknown model alias '$MODEL_ALIAS'"
+        show_help
+        ;;
+esac
+
+echo "Selected Model: $MODEL_ALIAS"
+echo "Repo ID: $MODEL_ID"
+echo "Target Directory: $DEFAULT_SAVE_DIR/$MODEL_ID"
+
+# Prepare environment for acceleration (Default to mirror if not set)
+if [ -z "$HF_ENDPOINT" ]; then
+    export HF_ENDPOINT=$MIRROR_URL
+    echo "Using default HF Mirror: $HF_ENDPOINT"
+else
+    echo "Using custom HF Endpoint: $HF_ENDPOINT"
+fi
+
+# Check if huggingface-cli and hf_transfer are installed
+CLI_CMD=""
+if command -v huggingface-cli &> /dev/null; then
+    CLI_CMD="huggingface-cli"
+elif command -v hf &> /dev/null; then
+    CLI_CMD="hf"
+else
+    echo "huggingface-cli not found. Installing via pip..."
+    if command -v pip3 &> /dev/null; then
+        pip3 install -U "huggingface_hub[cli]" hf_transfer
+    elif command -v pip &> /dev/null; then
+        pip install -U "huggingface_hub[cli]" hf_transfer
+    else
+        echo "Error: pip is not available. Please install python and pip."
+        exit 1
+    fi
+    CLI_CMD="huggingface-cli"
+fi
+
+# Try to install hf_transfer if simple python check fails (optional but recommended for speed)
+if ! python3 -c "import hf_transfer" &> /dev/null; then
+     echo "Installing hf_transfer for faster downloads..."
+     pip3 install -U hf_transfer || echo "Warning: hf_transfer install failed, falling back to standard download."
+fi
+
+# Enable HF Transfer (Rust-based downloader) - Default to 1 (On) unless explicitly disabled
+export HF_HUB_ENABLE_HF_TRANSFER=${HF_HUB_ENABLE_HF_TRANSFER:-1}
+if [ "$HF_HUB_ENABLE_HF_TRANSFER" == "1" ]; then
+    echo "Enabled HF_HUB_ENABLE_HF_TRANSFER for high-speed download"
+else
+    echo "HF_HUB_ENABLE_HF_TRANSFER disabled. Using standard Python downloader."
+fi
+
+# Download model
+echo "Downloading model with acceleration via $CLI_CMD..."
+# Note: --resume-download is deprecated/implicit in newer versions, removing it. 
+# --local-dir-use-symlinks matches standard CLI if version is recent. 
+if ! $CLI_CMD download "$MODEL_ID" --local-dir "$DEFAULT_SAVE_DIR/$MODEL_ID" --local-dir-use-symlinks False $TOKEN_ARG; then
+    echo ""
+    echo "Error: Download failed."
+    echo "If you received a 429 Rate Limit error, please provide a Hugging Face Token."
+    echo "Usage: ./download_and_run.sh $MODEL_ALIAS [hf_token]"
+    echo "Or set the HF_TOKEN environment variable."
+    exit 1
+fi
+
+# Run cargo
+echo "Starting application with cargo..."
+# Detect if on Mac to add 'metal' feature
+if [[ "$OSTYPE" == "darwin"* ]]; then
+    echo "Detected MacOS. Running with 'metal' feature..."
+    cargo run -r --features metal -- --model "$MODEL_ALIAS" --weight-path "$DEFAULT_SAVE_DIR/$MODEL_ID"
+else
+    echo "Running with default features (cuda)..."
+    cargo run -r --features cuda -- --model "$MODEL_ALIAS" --weight-path "$DEFAULT_SAVE_DIR/$MODEL_ID"
+fi

src/models/voxcpm/generate.rs

@@ -199,6 +199,10 @@ impl VoxCPMGenerate {
         )?;
         Ok(audio)
     }
+
+    pub fn sample_rate(&self) -> usize {
+        self.sample_rate
+    }
 }
 
 impl GenerateModel for VoxCPMGenerate {

tests/test_deepseek_ocr.rs

@@ -32,10 +32,11 @@ 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 +85,11 @@ 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 +8,8 @@ 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 +57,7 @@ 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_hunyuan_ocr.rs

@@ -31,10 +31,11 @@ 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 +80,11 @@ 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 +11,8 @@ 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 +28,7 @@ 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 +51,8 @@ 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 +67,7 @@ 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 +32,11 @@ 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 +80,11 @@ 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 +13,8 @@ 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 +41,7 @@ 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 +66,8 @@ 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 +94,7 @@ 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 +9,8 @@ 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 +37,7 @@ 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 +59,8 @@ 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 +87,7 @@ 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 +8,8 @@ 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 +32,7 @@ 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 +8,8 @@ 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 +29,7 @@ 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

@@ -10,10 +10,12 @@ use aha::{
 use aha_openai_dive::v1::resources::chat::ChatCompletionParameters;
 use anyhow::{Ok, Result};
 
+#[test]
 #[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 +61,11 @@ 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 +109,9 @@ fn voxcpm_generate() -> Result<()> {
 
 #[test]
 fn voxcpm_tokenizer() -> Result<()> {
-    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 tokenizer = SingleChineseTokenizer::new(model_path)?;
+    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 +13,8 @@ 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 +41,7 @@ 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 +59,11 @@ 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 +108,9 @@ 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 save_dir = aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
-    let tokenizer = SingleChineseTokenizer::new(model_path)?;
+    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 +7,9 @@ use candle_nn::VarBuilder;
 
 #[test]
 fn minicpm4_weight() -> Result<()> {
-    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_list = find_type_files(model_path, "safetensors")?;
+    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 +24,9 @@ fn minicpm4_weight() -> Result<()> {
 
 #[test]
 fn voxcpm_weight() -> Result<()> {
-    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_list = find_type_files(model_path, "pth")?;
+    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 +50,9 @@ fn voxcpm_weight() -> Result<()> {
 
 #[test]
 fn voxcpm1_5_weight() -> Result<()> {
-    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_list = find_type_files(model_path, "pth")?;
+    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 +71,9 @@ fn voxcpm1_5_weight() -> Result<()> {
 
 #[test]
 fn qwen3vl_weight() -> Result<()> {
-    let model_path = "/home/jhq/huggingface_model/Qwen/Qwen3-VL-4B-Instruct/";
+    let save_dir = aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
-    let model_list = find_type_files(model_path, "safetensors")?;
+    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 +88,9 @@ fn qwen3vl_weight() -> Result<()> {
 
 #[test]
 fn deepseekocr_weight() -> Result<()> {
-    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_list = find_type_files(model_path, "safetensors")?;
+    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 +108,9 @@ fn deepseekocr_weight() -> Result<()> {
 
 #[test]
 fn hunyuanocr_weight() -> 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_list = find_type_files(model_path, "safetensors")?;
+    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 {