refactor: reorganize imports and improve code formatting across multiple modules

src/exec/qwen3_asr.rs

@@ -5,14 +5,13 @@ use std::time::Instant;
 use anyhow::{Ok, Result};
 
 use crate::exec::ExecModel;
+use crate::models::GenerateModel;
 use crate::models::qwen3_asr::generate::Qwen3AsrGenerateModel;
-use crate::models::{GenerateModel};
 
 pub struct Qwen3ASRExec;
 
 impl ExecModel for Qwen3ASRExec {
     fn run(input: &[String], output: Option<&str>, weight_path: &str) -> Result<()> {
-        
         let i_start = Instant::now();
         let mut model = Qwen3AsrGenerateModel::init(weight_path, None, None)?;
         let i_duration = i_start.elapsed();

src/models/campplus/mod.rs

@@ -25,7 +25,11 @@ impl Shortcut {
     ) -> Result<Self> {
         let conv_0 = get_conv2d(vb.pp("0"), in_c, out_c, ks, padding, 1, 1, 1, bias)?;
         let bn_1 = get_batch_norm(vb.pp("1"), 1e-5, out_c, true)?;
-        Ok(Self { conv_0, bn_1, stride })
+        Ok(Self {
+            conv_0,
+            bn_1,
+            stride,
+        })
     }
 
     pub fn forward(&self, x: &Tensor) -> Result<Tensor> {

src/models/feature_extractor/mod.rs

@@ -1,3 +1,3 @@
-pub mod seamless_m4t_feature_extractor;
-pub mod feature_extraction_whisper;
 pub mod config;
+pub mod feature_extraction_whisper;
+pub mod seamless_m4t_feature_extractor;

src/models/glm_asr_nano/config.rs

@@ -11,8 +11,6 @@ pub struct GlmAsrNanoProcessorConfig {
     pub max_audio_len: usize,
 }
 
-
-
 #[derive(Debug, Clone, PartialEq, Deserialize)]
 pub struct GlmAsrNanoConfig {
     pub audio_config: GlmAsrAudioConfig,

src/models/glm_asr_nano/processor.rs

@@ -1,4 +1,3 @@
-
 use aha_openai_dive::v1::resources::chat::ChatCompletionParameters;
 use anyhow::Result;
 use candle_core::{D, DType, Device, IndexOp, Tensor};

src/models/index_tts2/processor.rs

@@ -5,13 +5,16 @@ use candle_nn::VarBuilder;
 
 use crate::{
     models::{
-        campplus::CAMPPlus, feature_extractor::seamless_m4t_feature_extractor::SeamlessM4TFeatureExtractor, index_tts2::config::{IndexTTS2Config, PreprocessParams}, mask_gct::model::RepCodec, w2v_bert_2_0::model::W2VBert2_0Model
+        campplus::CAMPPlus,
+        feature_extractor::seamless_m4t_feature_extractor::SeamlessM4TFeatureExtractor,
+        index_tts2::config::{IndexTTS2Config, PreprocessParams},
+        mask_gct::model::RepCodec,
+        w2v_bert_2_0::model::W2VBert2_0Model,
     },
     utils::{
         audio_utils::{
             create_hann_window, extract_audio_url, get_waveform_and_window_properties, kaldi_fbank,
-            kaldi_get_mel_banks, load_audio, mel_filter_bank, resample_simple, 
+            kaldi_get_mel_banks, load_audio, mel_filter_bank, resample_simple, torch_stft,
-            torch_stft,
         },
         get_vb_model_path,
         tensor_utils::pad_reflect_last_dim,

src/models/mask_gct/mod.rs

@@ -1,2 +1,2 @@
-pub mod model;
 pub mod config;
+pub mod model;

src/models/mask_gct/model.rs

@@ -116,10 +116,28 @@ impl FactorizedVectorQuantize {
         use_l2_normlize: bool,
     ) -> Result<Self> {
         let (in_project, out_project) = if input_dim != codebook_dim {
-            let in_project =
+            let in_project = WNConv1d::new(
-                WNConv1d::new(vb.pp("in_project"), input_dim, codebook_dim, 1, 1, 0, 1, 1, true)?;
+                vb.pp("in_project"),
-            let out_project =
+                input_dim,
-                WNConv1d::new(vb.pp("out_project"), codebook_dim, input_dim, 1, 1, 0, 1, 1, true)?;
+                codebook_dim,
+                1,
+                1,
+                0,
+                1,
+                1,
+                true,
+            )?;
+            let out_project = WNConv1d::new(
+                vb.pp("out_project"),
+                codebook_dim,
+                input_dim,
+                1,
+                1,
+                0,
+                1,
+                1,
+                true,
+            )?;
             (Some(in_project), Some(out_project))
         } else {
             (None, None)

src/models/qwen3_asr/generate.rs

@@ -1,4 +1,3 @@
-
 use aha_openai_dive::v1::resources::chat::{
     ChatCompletionChunkResponse, ChatCompletionParameters, ChatCompletionResponse,
 };

src/models/w2v_bert_2_0/model.rs

@@ -7,9 +7,7 @@ use candle_nn::{
 
 use crate::{
     models::{
-        common::{
+        common::{GLU, TwoLinearMLP, eager_attention_forward, get_conv1d, get_layer_norm},
-            GLU, TwoLinearMLP, eager_attention_forward, get_conv1d, get_layer_norm,
-        },
         w2v_bert_2_0::config::W2VBert2_0Config,
     },
     position_embed::rope::{RoPE, apply_rotary_pos_emb},

src/utils/audio_utils.rs

@@ -32,7 +32,9 @@ use symphonia::core::meta::MetadataOptions;
 use symphonia::core::probe::Hint;
 
 use crate::utils::get_default_save_dir;
-use crate::utils::tensor_utils::{linspace, log10, pad_reflect_last_dim, pad_replicate_last_dim, split_tensor};
+use crate::utils::tensor_utils::{
+    linspace, log10, pad_reflect_last_dim, pad_replicate_last_dim, split_tensor,
+};
 
 // 重采样方法枚举
 #[derive(Debug, Clone, Copy)]

tests/messy_test.rs

@@ -2,9 +2,9 @@
 
 use std::time::Instant;
 
-use aha::utils::{tensor_utils::interpolate_nearest_1d};
+use aha::utils::tensor_utils::interpolate_nearest_1d;
 use anyhow::Result;
-use candle_core::{Tensor};
+use candle_core::Tensor;
 // use symphonia::core::io::MediaSourceStream;
 
 #[test]
@@ -16,7 +16,10 @@ fn messy_test() -> Result<()> {
     let i_start = Instant::now();
     let t_inter = interpolate_nearest_1d(&t, 20)?;
     let i_duration = i_start.elapsed();
-    println!("Time elapsed in interpolate_nearest_1d is: {:?}", i_duration);
+    println!(
+        "Time elapsed in interpolate_nearest_1d is: {:?}",
+        i_duration
+    );
     println!("t_inter: {}", t_inter);
     // let url = "https://sis-sample-audio.obs.cn-north-1.myhuaweicloud.com/16k16bit.mp3";
     // let client = reqwest::blocking::Client::new();

tests/test_index_tts2.rs

@@ -1,8 +1,8 @@
 use std::time::Instant;
-use anyhow::Result;
 
 use aha::models::index_tts2::{generate::IndexTTS2Generate, utils::download_index_tts2_need_model};
 use aha_openai_dive::v1::resources::chat::ChatCompletionParameters;
+use anyhow::Result;
 
 #[tokio::test]
 async fn index_tts2_generate() -> Result<()> {