stash save

tests/config_tests.rs

@@ -1,13 +1,5 @@
 use aha::models::{
-    deepseek_ocr::config::DeepseekOCRConfig,
-    hunyuan_ocr::config::HunYuanVLConfig,
-    lfm2::config::Lfm2Config,
-    lfm2vl::config::{Lfm2ProcessorConfig, Lfm2VLConfig},
-    minicpm4::config::MiniCPM4Config,
-    paddleocr_vl::config::PaddleOCRVLConfig,
-    qwen2_5vl::config::Qwen2_5VLConfig,
-    qwen3vl::config::Qwen3VLConfig,
-    voxcpm::config::VoxCPMConfig,
+    deepseek_ocr::config::DeepseekOCRConfig, hunyuan_ocr::config::HunYuanVLConfig, lfm2::config::Lfm2Config, lfm2vl::config::{Lfm2ProcessorConfig, Lfm2VLConfig}, minicpm4::config::MiniCPM4Config, moss::config::{MossAudioTokenizerConfig, MossTTSConfig}, paddleocr_vl::config::PaddleOCRVLConfig, qwen2_5vl::config::Qwen2_5VLConfig, qwen3vl::config::Qwen3VLConfig, voxcpm::config::VoxCPMConfig
 };
 use anyhow::Result;
 
@@ -117,3 +109,23 @@ fn lfm2vl_config() -> Result<()> {
     println!("{:?}", processor_config);
     Ok(())
 }
+
+#[test]
+fn moss_audio_tokenizer_config() -> Result<()> {
+    // cargo test -F cuda --test config_tests moss_audio_tokenizer_config -r -- --nocapture
+    let model_path = "/home/jhq/.aha/openmoss/MOSS-Audio-Tokenizer-Nano/";
+    let config_path = model_path.to_string() + "/config.json";
+    let config: MossAudioTokenizerConfig = serde_json::from_slice(&std::fs::read(config_path)?)?;
+    println!("{:?}", config);
+    Ok(())
+}
+
+#[test]
+fn moss_tts_config() -> Result<()> {
+    // cargo test -F cuda --test config_tests moss_tts_config -r -- --nocapture
+    let model_path = "/home/jhq/.aha/openmoss/MOSS-TTS-Nano/";
+    let config_path = model_path.to_string() + "/config.json";
+    let config: MossTTSConfig = serde_json::from_slice(&std::fs::read(config_path)?)?;
+    println!("{:?}", config);
+    Ok(())
+}

tests/test_moss_tts.rs

@@ -0,0 +1,20 @@
+use aha::models::moss::generate::MossTTSGenerate;
+use anyhow::Result;
+
+#[test]
+fn moss_tts() -> Result<()> {
+    // RUST_BACKTRACE=1 cargo test -F cuda --test test_moss_tts moss_tts -r -- --nocapture
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let tts_path = format!("{}/openmoss/MOSS-TTS-Nano/", save_dir);
+    let audio_tokenizer_path = format!("{}/openmoss/MOSS-Audio-Tokenizer-Nano/", save_dir);
+    let mut model = MossTTSGenerate::init(&tts_path, &audio_tokenizer_path, None, None)?;
+    let _ = model.generate(
+        "您好啊,吃饭了吗,吃的啥啊中午",
+        Some("file://./assets/audio/jiangjiang.wav"),
+        Some("哈喽大家好，我是蒋蒋"),
+        Some(aha::models::moss::tts_nano::MossTTSMode::Continuation),
+        // None,
+    )?;
+    Ok(())
+}

tests/weight_test.rs

@@ -425,3 +425,30 @@ fn silero_vad_weight() -> Result<()> {
 
     Ok(())
 }
+
+#[test]
+fn moss_tts_nano_weight() -> Result<()> {
+    // cargo test -F cuda --test weight_test moss_tts_nano_weight -r -- --nocapture
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/openmoss/MOSS-TTS-Nano/pytorch_model.bin", save_dir);
+    let dict = read_all_with_key(&model_path, None)?;
+    for (k, v) in dict {
+        println!("key: {}, tensor shape: {:?}", k, v);
+    }
+    Ok(())
+}
+
+#[test]
+fn moss_audio_tokenizer_nano_weight() -> Result<()> {
+    // cargo test -F cuda --test weight_test moss_audio_tokenizer_nano_weight -r -- --nocapture
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/openmoss/MOSS-Audio-Tokenizer-Nano/model-00001-of-00001.safetensors", save_dir);
+    let device = get_device(None);
+    let weights = safetensors::load(model_path, &device)?;
+    for (key, tensor) in weights.iter() {
+        println!("=== {} === {:?}", key, tensor);
+    }
+    Ok(())
+}