add moss-tts-nano

Cargo.lock

@@ -885,7 +885,6 @@ dependencies = [
 ]
 
 [[package]]
-<<<<<<< HEAD
 name = "cpufeatures"
 version = "0.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
@@ -895,23 +894,6 @@ dependencies = [
 ]
 
 [[package]]
-name = "crc"
-version = "3.3.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9710d3b3739c2e349eb44fe848ad0b7c8cb1e42bd87ee49371df2f7acaf3e675"
-dependencies = [
- "crc-catalog",
-]
-
-[[package]]
-name = "crc-catalog"
-version = "2.4.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "19d374276b40fb8bbdee95aef7c7fa6b5316ec764510eb64b8dd0e2ed0d7e7f5"
-
-[[package]]
-=======
->>>>>>> main
 name = "crc32fast"
 version = "1.5.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"

src/exec/voxcpm.rs

@@ -51,7 +51,7 @@ impl ExecModel for VoxCPMExec {
         };
 
         let sample_rate = voxcpm_generate.sample_rate();
-        crate::utils::audio_utils::save_wav(&audio, &output_path, sample_rate as u32)?;
+        crate::utils::audio_utils::save_wav_mono(&audio, &output_path, sample_rate as u32)?;
 
         println!("Output saved to: {}", output_path);
 

src/models/common/generate.rs

@@ -1,12 +1,15 @@
 use anyhow::Result;
 use candle_core::{DType, Device, Tensor};
-use candle_transformers::generation::{LogitsProcessor};
+use candle_transformers::generation::LogitsProcessor;
 use rocket::async_stream::stream;
 use rocket::futures::Stream;
 use std::time::Instant;
 
 use crate::{
-    models::common::{InferenceModel, MultiModalData, sample::{use_repeat_penalty, get_logit_processor}},
+    models::common::{
+        InferenceModel, MultiModalData,
+        sample::{get_logit_processor, use_repeat_penalty},
+    },
     params::chat::{ChatCompletionChunkResponse, ChatCompletionResponse},
     tokenizer::TokenizerModel,
     utils::response_utils::{

src/models/common/modules.rs

@@ -1393,3 +1393,16 @@ pub fn quick_gelu(xs: &Tensor) -> Result<Tensor> {
     let x = sigmoid(&x)?;
     Ok(xs.mul(&x)?)
 }
+
+pub fn new_gelu(xs: &Tensor) -> Result<Tensor> {
+    let sqrt_two_over_pi = (2.0f64 / std::f64::consts::PI).sqrt();
+    Ok(xs
+        .powf(3.0)?
+        .affine(0.044715, 0.0)?
+        .add(xs)?
+        .affine(sqrt_two_over_pi, 0.0)?
+        .tanh()?
+        .affine(1.0, 1.0)?
+        .mul(xs)?
+        .affine(0.5, 0.0)?)
+}

src/models/common/sample.rs

@@ -2,7 +2,7 @@ use anyhow::{Result, anyhow};
 use candle_core::{IndexOp, Tensor};
 use candle_nn::ops::softmax;
 use candle_transformers::generation::{LogitsProcessor, Sampling};
-use rand::{SeedableRng, distr::Distribution};
+use rand::distr::Distribution;
 
 pub fn get_logit_processor(
     temperature: Option<f32>,
@@ -43,7 +43,7 @@ pub fn use_repeat_penalty(
     logits: &Tensor,
     context: &[u32],
 ) -> Result<Tensor> {
-    if repeat_penalty == 1.0 || repeat_last_n.map_or(false, |n| n == 0) {
+    if repeat_penalty == 1.0 || repeat_last_n == Some(0) {
         Ok(logits.clone())
     } else {
         let start_at = if let Some(last_n) = repeat_last_n {
@@ -52,13 +52,24 @@ pub fn use_repeat_penalty(
             0
         };
         Ok(candle_transformers::utils::apply_repeat_penalty(
-            &logits,
+            logits,
             repeat_penalty,
             &context[start_at..],
         )?)
     }
 }
 
+pub fn sample_weighted(prs: &[f32]) -> Result<u32> {
+    let mut rng = rand::rng();
+    let dist = rand::distr::weighted::WeightedIndex::new(prs).map_err(|e| {
+        anyhow!(format!(
+            "simple_sampel new  rand::distr::weighted::WeightedIndex Failed: {}",
+            e
+        ))
+    })?;
+    Ok(dist.sample(&mut rng) as u32)
+}
+
 /// logits shape: (dim)
 pub fn simple_sample(
     logits: &Tensor,
@@ -68,7 +79,6 @@ pub fn simple_sample(
     top_p: Option<f32>,
     previous_token_ids: Option<&[u32]>,
     repeat_penalty: f32,
-    seed: Option<u64>,
 ) -> Result<u32> {
     if logits.rank() != 1 {
         return Err(anyhow!("simple_sample logits need rank = 1"));
@@ -90,7 +100,7 @@ pub fn simple_sample(
         }
         if let Some(top_k) = top_k
             && top_k > 0
-            && top_k > logits.dim(0)?
+            && top_k < logits.dim(0)?
         {
             let sorted_indices = logits.arg_sort_last_dim(false)?;
             let top_k_indices = sorted_indices.narrow(0, 0, top_k)?;
@@ -114,7 +124,7 @@ pub fn simple_sample(
                 .broadcast_gt(&Tensor::new(top_p, logits.device())?.to_dtype(logits.dtype())?)?;
             // 保证数据不会被全部置为-inf
             if mask.i(0)?.to_scalar::<u8>()? == 1 {
-                mask = mask.slice_scatter(&Tensor::new(0u32, logits.device())?, 0, 0)?;
+                mask = mask.slice_scatter(&Tensor::new(&[0u8], logits.device())?, 0, 0)?;
             }
             let on_true = Tensor::new(f32::NEG_INFINITY, logits.device())?
                 .to_dtype(logits.dtype())?
@@ -122,19 +132,9 @@ pub fn simple_sample(
             let new_logits = mask.where_cond(&on_true, &sorted_logits)?;
             logits = logits.scatter(&sorted_indices, &new_logits, 0)?;
         }
+        let probs = softmax(&logits, 0)?;
 
-        let probs = softmax(&logits, 0)?
+        let probs = probs.to_dtype(candle_core::DType::F32)?.to_vec1::<f32>()?;
-            .to_dtype(candle_core::DType::F32)?
+        sample_weighted(&probs)
-            .to_vec1::<f32>()?;
-        let distr = rand::distr::weighted::WeightedIndex::new(probs).map_err(|e| {
-            anyhow!(format!(
-                "simple_sampel new  rand::distr::weighted::WeightedIndex Failed: {}",
-                e
-            ))
-        })?;
-        let seed = seed.unwrap_or(34567);
-        let mut rng = rand::rngs::StdRng::seed_from_u64(seed);
-        let next_token = distr.sample(&mut rng) as u32;
-        Ok(next_token)
     }
 }

src/models/glm_asr_nano/model.rs

@@ -6,11 +6,10 @@ use crate::{
     models::{
         common::{
             InferenceModel,
-            modules::{
+            modules::{TwoLinearMLP, eager_attention_forward, get_conv1d, get_layer_norm},
-                TwoLinearMLP, eager_attention_forward, get_conv1d, get_layer_norm,
-            },
         },
-        glm_asr_nano::config::{GlmAsrAudioConfig, GlmAsrNanoConfig}, llama::LlamaForCausalLM,
+        glm_asr_nano::config::{GlmAsrAudioConfig, GlmAsrNanoConfig},
+        llama::LlamaForCausalLM,
     },
     position_embed::rope::{RoPE, glm_asr_apply_rotary_pos_emb},
     utils::tensor_utils::{get_equal_mask, masked_scatter_dim0},

src/models/gpt2/mod.rs

@@ -71,7 +71,7 @@ impl GPT2Attention {
             }
         };
         self.kv_cache = Some((key_states.clone(), value_states.clone()));
-        let scale = 1f64 / f64::sqrt(self.head_dim as f64);
+        let scale: f64 = 1f64 / f64::sqrt(self.head_dim as f64);
         let attn_output = eager_attention_forward(
             &query_states,
             &key_states,
@@ -277,7 +277,8 @@ impl GPT2Model {
     pub fn forward(&mut self, inputs_embeds: &Tensor, seqlen_offset: usize) -> Result<Tensor> {
         let (b_size, seq_len, _) = inputs_embeds.dims3()?;
         let (cos, sin) = if let Some(rope) = &self.rope {
-            let (cos, sin) = rope.forward_repeat_interleave(seqlen_offset, seq_len, inputs_embeds.device())?;
+            let (cos, sin) =
+                rope.forward_repeat_interleave(seqlen_offset, seq_len, inputs_embeds.device())?;
             (Some(cos), Some(sin))
         } else {
             (None, None)

src/models/mod.rs

@@ -12,7 +12,8 @@ pub mod lfm2;
 pub mod lfm2vl;
 pub mod mask_gct;
 pub mod minicpm4;
-pub mod moss;
+pub mod moss_audio_tokenizer_nano;
+pub mod moss_tts_nano;
 pub mod paddleocr_vl;
 pub mod qwen2;
 pub mod qwen2_5vl;

src/models/moss_audio_tokenizer_nano/config.rs

@@ -0,0 +1,53 @@
+use serde::Deserialize;
+
+#[derive(Debug, Deserialize)]
+pub struct MossAudioTokenizerConfig {
+    pub sample_rate: usize,
+    pub sampling_rate: usize,
+    pub downsample_rate: usize,
+    pub causal_transformer_context_duration: f64,
+    pub number_channels: usize,
+    pub enable_channel_interleave: bool,
+    pub compute_dtype: String,
+    pub dtype: String,
+    pub code_dim: usize,
+    pub encoder_kwargs: Vec<MossAudioTokenizerModuleConfig>,
+    pub decoder_kwargs: Vec<MossAudioTokenizerModuleConfig>,
+    pub quantizer_type: String,
+    pub quantizer_kwargs: MossAudioTokenizerQuantizerKwargs,
+    pub reversed_decoder_kwargs: Vec<MossAudioTokenizerModuleConfig>,
+}
+
+#[derive(Debug, Deserialize)]
+pub struct MossAudioTokenizerModuleConfig {
+    pub module_type: String,
+    pub patch_size: Option<usize>,
+    pub causal: Option<bool>,
+    pub context_duration: Option<f64>,
+    pub conv_layout: Option<bool>,
+    pub d_model: Option<usize>,
+    pub dim_feedforward: Option<usize>,
+    pub gating: Option<String>,
+    pub input_dimension: Option<usize>,
+    pub layer_scale: Option<f64>,
+    pub max_period: Option<usize>,
+    pub norm: Option<String>,
+    pub num_heads: Option<usize>,
+    pub num_layers: Option<usize>,
+    pub output_dimension: Option<usize>,
+    pub positional_embedding: Option<String>,
+}
+
+#[derive(Debug, Deserialize)]
+pub struct MossAudioTokenizerQuantizerKwargs {
+    pub codebook_dim: usize,
+    pub codebook_loss_weight: f64,
+    pub codebook_size: usize,
+    pub commitment_loss_weight: f64,
+    pub input_dim: usize,
+    pub num_quantizers: usize,
+    pub output_dim: usize,
+    pub quantizer_dropout: f64,
+    pub quantizer_type: String,
+    pub rvq_dim: usize,
+}

src/models/moss_audio_tokenizer_nano/mod.rs

@@ -1,3 +1,4 @@
+pub mod config;
 use anyhow::{Result, anyhow};
 use candle_core::{D, IndexOp, Tensor};
 use candle_nn::{Embedding, LayerNorm, Linear, Module, VarBuilder, embedding, linear_no_bias};
@@ -7,7 +8,7 @@ use crate::{
         common::modules::{
             TwoLinearMLP, WNConv1d, eager_attention_forward, get_layer_norm, l2_normalize,
         },
-        moss::config::{
+        moss_audio_tokenizer_nano::config::{
             MossAudioTokenizerConfig, MossAudioTokenizerModuleConfig,
             MossAudioTokenizerQuantizerKwargs,
         },
@@ -47,7 +48,8 @@ impl MossAudioTokenizerPatchedPretransform {
             .reshape((b, d, self.patch_size, l))?
             .permute((0, 1, 3, 2))?
             .reshape((b, d, l * self.patch_size))?;
-        let out_lengths = (input_lengths * self.patch_size as f64)?;
+        // let out_lengths = (input_lengths * self.patch_size as f64)?;
+        let out_lengths = input_lengths.affine(self.patch_size as f64, 0.0)?;
         Ok((x, out_lengths))
     }
 
@@ -398,12 +400,21 @@ impl MossAudioTokenizerLFQ {
         }
         Ok((z_q, indices))
     }
+
+    pub fn decode_code(&self, codec: &Tensor) -> Result<Tensor> {
+        let mut z_q = self.codebook.forward(codec)?.transpose(1, 2)?;
+        if let Some(out_proj) = &self.out_proj {
+            z_q = out_proj.forward(&z_q)?;
+        }
+        Ok(z_q)
+    }
 }
 
 pub struct MossAudioTokenizerResidualLFQ {
     input_proj: Option<WNConv1d>,
     output_proj: Option<WNConv1d>,
     quantizers: Vec<MossAudioTokenizerLFQ>,
+    rvq_dim: usize,
 }
 
 impl MossAudioTokenizerResidualLFQ {
@@ -454,6 +465,7 @@ impl MossAudioTokenizerResidualLFQ {
             input_proj,
             output_proj,
             quantizers,
+            rvq_dim: config.rvq_dim,
         })
     }
 
@@ -483,6 +495,23 @@ impl MossAudioTokenizerResidualLFQ {
         let all_indices = Tensor::stack(&all_indices, 0)?;
         Ok(all_indices)
     }
+
+    pub fn decode_codes(&self, codes: &Tensor) -> Result<Tensor> {
+        let (_, bs, t) = codes.dims3()?;
+        let mut emb = Tensor::zeros(
+            (bs, self.rvq_dim, t),
+            candle_core::DType::F32,
+            codes.device(),
+        )?;
+        for (i, quantizer) in self.quantizers.iter().enumerate() {
+            let code_i = quantizer.decode_code(&codes.i(i)?)?;
+            emb = emb.add(&code_i)?;
+        }
+        if let Some(output_proj) = &self.output_proj {
+            emb = output_proj.forward(&emb)?;
+        }
+        Ok(emb)
+    }
 }
 
 pub struct MossAudioTokenizer {
@@ -538,7 +567,7 @@ impl MossAudioTokenizer {
             if cfg.module_type == "PatchedPretransform"
                 && let Some(patch_size) = cfg.patch_size
             {
-                let layer = MossAudioTokenizerPatchedPretransform::new(patch_size, true);
+                let layer = MossAudioTokenizerPatchedPretransform::new(patch_size, false);
                 decoder.push(MossAudioTokenizerModule::PatchedPretransform(layer));
             } else if cfg.module_type == "Transformer" {
                 let context_duration = cfg
@@ -617,7 +646,7 @@ impl MossAudioTokenizer {
     }
 
     pub fn encode_one(&self, wav: &Tensor) -> Result<Tensor> {
-        // (channel, audio_len) -> (bs=1, channel, audio_len)
+        // in: (channel, audio_len)
         let (c, len) = wav.dims2()?;
         if c != self.number_channels {
             return Err(anyhow!(
@@ -632,7 +661,7 @@ impl MossAudioTokenizer {
         Ok(audio_vec[0].clone())
     }
 
-    pub fn encode_list(&self, wavs: &Vec<Tensor>) -> Result<Vec<Tensor>> {
+    pub fn encode_list(&self, wavs: &[Tensor]) -> Result<Vec<Tensor>> {
         if wavs.is_empty() {
             return Err(anyhow!(
                 "MossAudioTokenizer encode_list need wavs len > 0, but the wavs is empty"
@@ -664,6 +693,27 @@ impl MossAudioTokenizer {
         let input_values = Tensor::stack(&input_values, 0)?;
         let length_tensor = Tensor::new(length.clone(), input_values.device())?
             .to_dtype(candle_core::DType::F32)?;
-        Ok(self.batch_encode(&input_values, &length_tensor)?)
+        self.batch_encode(&input_values, &length_tensor)
+    }
+
+    pub fn decode_audio_token_ids_to_waveform(&self, audio_token_ids: &Tensor) -> Result<Tensor> {
+        let decode_codes = audio_token_ids.t()?.unsqueeze(1)?; //(len, nq) -> (nq, len) -> (nq, 1, len)
+        let len = decode_codes.dim(2)?;
+        let mut audio = self.quantizer.decode_codes(&decode_codes)?;
+        let mut audio_length = Tensor::new(&[len as f32], audio.device())?;
+        for decoder_module in self.decoder.iter() {
+            (audio, audio_length) = decoder_module.forward(&audio, &audio_length)?;
+        }
+        if self.number_channels == 1 || !self.enable_channel_interleave {
+            Ok(audio)
+        } else {
+            let bs = audio.dim(0)?;
+            Ok(audio
+                .squeeze(1)?
+                .contiguous()?
+                .reshape((bs, (), self.number_channels))?
+                .transpose(1, 2)?
+                .contiguous()?)
+        }
     }
 }

src/models/moss_tts_nano/config.rs

@@ -2,59 +2,6 @@ use serde::Deserialize;
 
 use crate::models::gpt2::config::GPT2Config;
 
-#[derive(Debug, Deserialize)]
-pub struct MossAudioTokenizerConfig {
-    pub sample_rate: usize,
-    pub sampling_rate: usize,
-    pub downsample_rate: usize,
-    pub causal_transformer_context_duration: f64,
-    pub number_channels: usize,
-    pub enable_channel_interleave: bool,
-    pub compute_dtype: String,
-    pub dtype: String,
-    pub code_dim: usize,
-    pub encoder_kwargs: Vec<MossAudioTokenizerModuleConfig>,
-    pub decoder_kwargs: Vec<MossAudioTokenizerModuleConfig>,
-    pub quantizer_type: String,
-    pub quantizer_kwargs: MossAudioTokenizerQuantizerKwargs,
-    pub reversed_decoder_kwargs: Vec<MossAudioTokenizerModuleConfig>,
-}
-
-
-#[derive(Debug, Deserialize)]
-pub struct MossAudioTokenizerModuleConfig {
-    pub module_type: String,
-    pub patch_size: Option<usize>,
-    pub causal: Option<bool>,
-    pub context_duration: Option<f64>,
-    pub conv_layout: Option<bool>,
-    pub d_model: Option<usize>,
-    pub dim_feedforward: Option<usize>,
-    pub gating: Option<String>,
-    pub input_dimension: Option<usize>,
-    pub layer_scale: Option<f64>,
-    pub max_period: Option<usize>,
-    pub norm: Option<String>,
-    pub num_heads: Option<usize>,
-    pub num_layers: Option<usize>,
-    pub output_dimension: Option<usize>,
-    pub positional_embedding: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-pub struct MossAudioTokenizerQuantizerKwargs {
-    pub codebook_dim: usize,
-    pub codebook_loss_weight: f64,
-    pub codebook_size: usize,
-    pub commitment_loss_weight: f64,
-    pub input_dim: usize,
-    pub num_quantizers: usize,
-    pub output_dim: usize,
-    pub quantizer_dropout: f64,
-    pub quantizer_type: String,
-    pub rvq_dim: usize,
-}
-
 #[derive(Debug, Deserialize)]
 pub struct MossTTSConfig {
     pub add_cross_attention: bool,
@@ -67,7 +14,7 @@ pub struct MossTTSConfig {
     pub audio_tokenizer_sample_rate: usize,
     pub audio_user_slot_token_id: u32,
     pub audio_vocab_size: usize,
-    
+
     // Generation/Model Params (Simplified nullables to Options or defaults if not critical)
     pub bad_words_ids: Option<Vec<u32>>,
     pub begin_suppress_tokens: Option<Vec<u32>>,
@@ -85,75 +32,71 @@ pub struct MossTTSConfig {
     pub finetuning_task: Option<String>,
     pub forced_bos_token_id: Option<u32>,
     pub forced_eos_token_id: Option<u32>,
-    
+
     // GPT2 Backbone Config
     pub gpt2_config: GPT2Config,
-    
+
     pub hidden_size: usize,
     pub id2label: std::collections::HashMap<usize, String>,
-    
+
     pub im_end_token_id: u32,
     pub im_start_token_id: u32,
     pub initializer_range: f64,
     pub is_decoder: bool,
     pub is_encoder_decoder: bool,
     pub label2id: std::collections::HashMap<String, usize>,
-    
+
     pub length_penalty: f64,
     pub local_transformer_attn_implementation: String,
     pub local_transformer_layers: usize,
-    
+
     pub max_length: usize,
     pub max_position_embeddings: usize,
     pub min_length: usize,
-    
+
     pub model_architecture: String,
     pub model_type: String,
-    
+
     pub n_vq: usize,
     pub no_repeat_ngram_size: usize,
-    
+
     pub num_beam_groups: usize,
     pub num_beams: usize,
     pub num_return_sequences: usize,
-    
+
     pub output_attentions: bool,
     pub output_hidden_states: bool,
     pub output_scores: bool,
-    
+
     pub pad_token_id: u32,
     pub prefix: Option<String>,
     pub problem_type: Option<String>,
     // pub pruned_heads: std::collections::HashMap<String, Vec<usize>>,
-    
     pub remove_invalid_values: bool,
     pub repetition_penalty: f64,
-    
+
     pub return_dict: bool,
     pub return_dict_in_generate: bool,
-    
+
     pub sep_token_id: Option<u32>,
     pub suppress_tokens: Option<Vec<u32>>,
-    
+
     pub task_specific_params: Option<serde_json::Value>,
-    
+
     pub temperature: f32,
     pub tf_legacy_loss: bool,
     pub tie_encoder_decoder: bool,
     pub tie_word_embeddings: bool,
-    
+
     pub tokenizer_class: String,
     pub tokenizer_use_fast: bool,
-    
+
     pub top_k: usize,
     pub top_p: f32,
     pub torchscript: bool,
-    
+
     pub typical_p: f64,
-    
+
     pub use_bfloat16: bool,
     pub vocab_size: usize,
-    
 }
-
-

src/models/moss_tts_nano/generate.rs

@@ -1,11 +1,13 @@
 use std::collections::HashMap;
 
 use crate::{
-    models::moss::{
+    models::{
-        audio_tokenizer_nano::MossAudioTokenizer,
+        moss_audio_tokenizer_nano::{MossAudioTokenizer, config::MossAudioTokenizerConfig},
-        config::{MossAudioTokenizerConfig, MossTTSConfig},
+        moss_tts_nano::{
-        processor::MossTTSProcessor,
+            config::MossTTSConfig,
-        tts_nano::{MossTTSMode, MossTTSModel},
+            model::{MossTTSMode, MossTTSModel},
+            processor::MossTTSProcessor,
+        },
     },
     utils::{find_type_files, get_device, get_dtype},
 };
@@ -33,7 +35,7 @@ impl MossTTSGenerate {
         let audio_tokenizer_cfg: MossAudioTokenizerConfig =
             serde_json::from_slice(&std::fs::read(audio_tokenizer_config_path)?)?;
         let model_list = find_type_files(audio_tokenizer_path, "safetensors")?;
-        let audio_dtype = get_dtype(dtype.clone(), &audio_tokenizer_cfg.dtype);
+        let audio_dtype = get_dtype(dtype, &audio_tokenizer_cfg.dtype);
         let device = get_device(device);
         let vb = unsafe { VarBuilder::from_mmaped_safetensors(&model_list, audio_dtype, &device)? };
         let audio_tokenizer = MossAudioTokenizer::new(vb, &audio_tokenizer_cfg)?;
@@ -50,12 +52,10 @@ impl MossTTSGenerate {
         )?;
         let model_list = find_type_files(tts_path, "bin")?;
         let mut dict_to_hashmap = HashMap::new();
-        // let cfg_dtype = tts_cfg.dtype.as_str();
         let m_dtype = get_dtype(dtype, "bfloat16");
         for m in model_list {
             let dict = read_all_with_key(m, None)?;
             for (k, v) in dict {
-                // println!("key: {}, tensor shape: {:?}", k, v);
                 dict_to_hashmap.insert(k, v);
             }
         }
@@ -78,18 +78,21 @@ impl MossTTSGenerate {
         prompt_text: Option<&str>,
         mode: Option<MossTTSMode>,
     ) -> Result<()> {
-        let (mut input_ids, mask) = self.processor.build_inference_input_ids(
+        let mode = self.processor.resolved_mode(
+            mode,
+            prompt_text.is_some(),
+            prompt_audio_path.is_some(),
+        )?;
+        let input_ids = self.processor.build_inference_input_ids(
             text,
             prompt_audio_path,
             prompt_text,
-            mode,
+            mode.clone(),
             &self.audio_tokenizer,
             &self.text_tokenizer,
             &self.device,
         )?;
-        let _ = self.model.generate(&input_ids, Some(&mask))?;
+        self.model.generate(&input_ids, &self.audio_tokenizer)?;
-        // println!("input_ids: {}", input_ids);
-        // println!("mask: {}", mask);
         Ok(())
     }
 }

src/models/moss_tts_nano/mod.rs

@@ -1,5 +1,4 @@
-pub mod audio_tokenizer_nano;
 pub mod config;
 pub mod generate;
+pub mod model;
 pub mod processor;
-pub mod tts_nano;

src/models/moss_tts_nano/model.rs

@@ -1,9 +1,16 @@
-use crate::models::{common::sample::simple_sample, gpt2::GPT2Model, moss::config::MossTTSConfig};
+use crate::{
+    models::{
+        common::sample::simple_sample, gpt2::GPT2Model,
+        moss_audio_tokenizer_nano::MossAudioTokenizer, moss_tts_nano::config::MossTTSConfig,
+    },
+    utils::audio_utils::save_wav,
+};
 use anyhow::{Result, anyhow};
 use candle_core::{D, IndexOp, Tensor};
 use candle_nn::{Embedding, Linear, Module, VarBuilder, embedding, linear_no_bias};
+// use candle_transformers::generation::LogitsProcessor;
 
-#[derive(PartialEq, Debug)]
+#[derive(PartialEq, Clone, Debug)]
 pub enum MossTTSMode {
     Continuation,
     VoiceClone,
@@ -24,6 +31,7 @@ pub struct MossTTSModel {
     audio_top_k: usize,
     audio_top_p: f32,
     audio_repetition_penalty: f32,
+    // audio_processor: LogitsProcessor,
 }
 
 impl MossTTSModel {
@@ -92,6 +100,7 @@ impl MossTTSModel {
             audio_top_k: 25,
             audio_top_p: 0.95,
             audio_repetition_penalty: 1.2,
+            // audio_processor,
         })
     }
 
@@ -144,15 +153,9 @@ impl MossTTSModel {
             .i(self.audio_end_token_id)?
             .to_dtype(candle_core::DType::F32)?
             .to_scalar::<f32>()?;
-        println!(
+        let logits = Tensor::new(&[slot_token_id_logit, end_token_id_logit], logits.device())?;
-            "slot_token_id: {} logit: {slot_token_id_logit}",
+        let token = simple_sample(&logits, true, None, None, None, None, 1.0)?;
-            self.audio_assistant_slot_token_id
+        if token == 0 {
-        );
-        println!(
-            "end_token_id: {} logit: {end_token_id_logit}",
-            self.audio_end_token_id
-        );
-        if slot_token_id_logit > end_token_id_logit {
             Ok(self.audio_assistant_slot_token_id)
         } else {
             Ok(self.audio_end_token_id)
@@ -169,31 +172,28 @@ impl MossTTSModel {
         Ok(Tensor::cat(&[&slot, &audio_token_ids], D::Minus1)?)
     }
 
-    pub fn generate(&mut self, input_ids: &Tensor, mask: Option<&Tensor>) -> Result<()> {
+    pub fn generate(
-        let sample_len = 2;
+        &mut self,
+        input_ids: &Tensor,
+        audio_tokenizer: &MossAudioTokenizer,
+    ) -> Result<()> {
+        let sample_len = 100;
         let mut seqlen_offset = 0;
         let mut seq_len = input_ids.dim(1)?;
         let mut generated_frames = vec![];
         let mut current_model_input_ids = input_ids.clone();
-        for step_index in 0..sample_len {
+        for _ in 0..sample_len {
-            // println!("current_model_input_ids: {:?}", current_model_input_ids);
             let inputs_embeds = self.build_inputs_embeds(&current_model_input_ids)?;
             let outputs = self.transformer.forward(&inputs_embeds, seqlen_offset)?;
-            // println!("transformer-----------------------");
             let outputs_len = outputs.dim(1)?;
             let global_hidden_state = outputs.narrow(1, outputs_len - 1, 1)?;
-            // println!("global_hidden_state: {}", global_hidden_state);
             let mut local_positions = 0usize;
             let local_outputs = self
                 .local_transformer
                 .forward(&global_hidden_state, local_positions)?;
-            // println!("local_outputs-----------------------");
             let local_len = local_outputs.dim(1)?;
             let local_hidden_states = local_outputs.narrow(1, local_len - 1, 1)?;
-            // println!("local_hidden_states: {}", local_hidden_states);
             let text_logits = self.text_lm_head.forward(&local_hidden_states)?;
-            // println!("text_logits: {}", text_logits.i((0, 0, 0..100))?);
-            println!("step_index: {}", step_index);
             let next_text_token = self.sample_next_assistant_text_token(&text_logits)?;
             if next_text_token == self.audio_end_token_id {
                 self.local_transformer.clear_kv_cache();
@@ -216,14 +216,11 @@ impl MossTTSModel {
                     .forward(&current_local_input, local_positions)?;
                 let local_len = local_outputs.dim(1)?;
                 let local_hidden_states = local_outputs.narrow(1, local_len - 1, 1)?;
-                // println!("local_hidden_states: {local_hidden_states}");
+                let channel_logits = self.audio_lm_heads[channel_index]
-                let channel_logits = (&self.audio_lm_heads[channel_index])
                     .forward(&local_hidden_states)?
                     .squeeze(0)?
                     .squeeze(0)?;
-                // println!("channel_logits: {}", channel_logits.i(0..100)?);
+                // let channel_token = self.audio_processor.sample(&channel_logits)?;
-                let arg_max = channel_logits.argmax(0)?;
-                println!("arg_max: {}", arg_max);
                 let channel_token = simple_sample(
                     &channel_logits,
                     true,
@@ -232,25 +229,48 @@ impl MossTTSModel {
                     Some(self.audio_top_p),
                     Some(&next_frame_tokens),
                     self.audio_repetition_penalty,
-                    None,
                 )?;
-                println!("channel_token: {channel_token}");
                 next_frame_tokens.push(channel_token);
-                current_local_input = (&self.audio_embeddings[channel_index]).forward(
+                current_local_input = self.audio_embeddings[channel_index].forward(
                     &Tensor::from_slice(&[channel_token], (1, 1), input_ids.device())?,
                 )?;
-                // println!("current_local_input: {current_local_input}");
             }
             self.local_transformer.clear_kv_cache();
             let next_frame = Tensor::new(next_frame_tokens, input_ids.device())?;
-            // println!("next_frame: {next_frame}");
             current_model_input_ids = self.build_generation_row(&next_frame)?;
             seqlen_offset += seq_len;
             seq_len = 1;
             generated_frames.push(next_frame);
         }
         let audio_token_ids = Tensor::stack(&generated_frames, 0)?;
-        println!("audio_token_ids: {audio_token_ids}");
+        let waveform = audio_tokenizer
+            .decode_audio_token_ids_to_waveform(&audio_token_ids)?
+            .squeeze(0)?;
+        save_wav(
+            &waveform,
+            "./demo.wav",
+            2,
+            audio_tokenizer.sampling_rate as u32,
+        )?;
+        Ok(())
+    }
+
+    pub fn decode(
+        &self,
+        prompt_audio_code: Option<&Tensor>,
+        audio_tokenizer: &MossAudioTokenizer,
+    ) -> Result<()> {
+        if let Some(audio) = prompt_audio_code {
+            let waveform = audio_tokenizer
+                .decode_audio_token_ids_to_waveform(audio)?
+                .squeeze(0)?;
+            save_wav(
+                &waveform,
+                "./demo.wav",
+                2,
+                audio_tokenizer.sampling_rate as u32,
+            )?;
+        }
         Ok(())
     }
 }

src/models/moss_tts_nano/processor.rs

@@ -1,6 +1,7 @@
 use crate::{
-    models::moss::{
+    models::{
-        audio_tokenizer_nano::MossAudioTokenizer, config::MossTTSConfig, tts_nano::MossTTSMode,
+        moss_audio_tokenizer_nano::MossAudioTokenizer,
+        moss_tts_nano::{config::MossTTSConfig, model::MossTTSMode},
     },
     tokenizer::sentencepiece_encode_vec,
     utils::{audio_utils::load_audio_with_resample, prepare_tts_text},
@@ -70,7 +71,7 @@ impl MossTTSProcessor {
         })
     }
 
-    fn resolved_mode(
+    pub fn resolved_mode(
         &self,
         mode: Option<MossTTSMode>,
         has_prompt_text: bool,
@@ -99,12 +100,11 @@ impl MossTTSProcessor {
         text: &str,
         prompt_audio_path: Option<&str>,
         prompt_text: Option<&str>,
-        mode: Option<MossTTSMode>,
+        mode: MossTTSMode,
         audio_tokenizer: &MossAudioTokenizer,
         text_tokenizer: &SentencePieceProcessor,
         device: &Device,
-    ) -> Result<(Tensor, Tensor)> {
+    ) -> Result<Tensor> {
-        let mode = self.resolved_mode(mode, prompt_text.is_some(), prompt_audio_path.is_some())?;
         let audio_code = if let Some(audio_path) = prompt_audio_path {
             let audio = load_audio_with_resample(
                 audio_path,
@@ -142,7 +142,7 @@ impl MossTTSProcessor {
             suffix_token_ids.extend_from_slice(&self.assistant_ids);
             suffix_token_ids.push(self.audio_start_token_id);
             let audio_prefix_rows = Self::build_audio_prefix_rows(
-                &prompt_audio_codes,
+                prompt_audio_codes,
                 self.audio_user_slot_token_id,
                 device,
             )?;
@@ -155,9 +155,7 @@ impl MossTTSProcessor {
             let input_ids =
                 Tensor::cat(&[&prompt_ids_tensor, &audio_prefix_rows, &suffix_rows], 0)?
                     .unsqueeze(0)?;
-            let (bs, len, _) = input_ids.dims3()?;
+            Ok(input_ids)
-            let mask = Tensor::ones((bs, len), candle_core::DType::U32, device)?;
-            Ok((input_ids, mask))
         } else {
             let text = if let Some(prompt_text) = prompt_text {
                 prompt_text + text
@@ -176,20 +174,17 @@ impl MossTTSProcessor {
                 Self::build_text_raw(&prompt_ids, self.audio_pad_token_id, self.n_vq, device)?;
             if let Some(prompt_audio_codes) = &audio_code {
                 let audio_prefix_rows = Self::build_audio_prefix_rows(
-                    &prompt_audio_codes,
+                    prompt_audio_codes,
                     self.audio_assistant_slot_token_id,
                     device,
                 )?;
                 input_ids = Tensor::cat(&[&input_ids, &audio_prefix_rows], 0)?;
             }
             input_ids = input_ids.unsqueeze(0)?;
-            let (bs, len, _) = input_ids.dims3()?;
+            Ok(input_ids)
-            let mask = Tensor::ones((bs, len), candle_core::DType::U32, device)?;
-            Ok((input_ids, mask))
         }
     }
 
-
     fn build_audio_prefix_rows(
         prompt_audio_codes: &Tensor,
         slot_token_id: u32,
@@ -202,7 +197,7 @@ impl MossTTSProcessor {
     }
 
     fn build_text_raw(
-        token_ids: &Vec<u32>,
+        token_ids: &[u32],
         audio_pad_token_id: u32,
         n_vq: usize,
         device: &Device,

src/models/qwen3_asr/generate.rs

@@ -4,7 +4,8 @@ use crate::{
     models::common::{
         MultiModalData,
         generate::{GenerationContext, generate_generic_text},
-        modules::{AsrResult, VadFrameResult}, sample::get_logit_processor,
+        modules::{AsrResult, VadFrameResult},
+        sample::get_logit_processor,
     },
     params::chat::{ChatCompletionChunkResponse, ChatCompletionParameters, ChatCompletionResponse},
     utils::response_utils::{build_chunk_response_with_usage, build_completion_response_with_time},

src/utils/audio_utils.rs

@@ -647,7 +647,8 @@ pub fn load_audio_with_resample(
     resample_audio_from_bytes(audio_vec, device, target_sample_rate, target_channels)
 }
 
-pub fn save_wav(audio: &Tensor, save_path: &str, sample_rate: u32) -> Result<()> {
+/// audio: (1, len)
+pub fn save_wav_mono(audio: &Tensor, save_path: &str, sample_rate: u32) -> Result<()> {
     let spec = hound::WavSpec {
         channels: 1,
         sample_rate,
@@ -669,6 +670,49 @@ pub fn save_wav(audio: &Tensor, save_path: &str, sample_rate: u32) -> Result<()>
     Ok(())
 }
 
+/// audio: (channel, len)
+pub fn save_wav(audio: &Tensor, save_path: &str, channels: usize, sample_rate: u32) -> Result<()> {
+    if audio.rank() != 2 {
+        return Err(anyhow!(
+            "Audio tensor must be 2D (channels, len), got shape {:?}",
+            audio.dims()
+        ));
+    }
+
+    let (num_channels, len) = audio.dims2()?;
+    if num_channels != channels {
+        return Err(anyhow!(
+            "Channel mismatch: Tensor has {} channels, but spec requires {}",
+            num_channels,
+            channels
+        ));
+    }
+    let spec = hound::WavSpec {
+        channels: channels as u16,
+        sample_rate,
+        bits_per_sample: 16,
+        sample_format: hound::SampleFormat::Int,
+    };
+
+    let max_val = audio.abs()?.max_all()?.to_scalar::<f32>()?;
+    let ratio = if max_val > 1.0 {
+        32767.0 / max_val
+    } else {
+        32767.0
+    };
+    let audio_vec_2d = audio.to_vec2::<f32>()?;
+    let mut writer = hound::WavWriter::create(save_path, spec).unwrap();
+    for i in 0..len {
+        for ch in 0..num_channels {
+            let sample_i16 = (audio_vec_2d[ch][i] * ratio).round() as i16;
+            let sample_i16 = sample_i16.clamp(-32768, 32767);
+            writer.write_sample(sample_i16)?;
+        }
+    }
+    writer.finalize()?;
+    Ok(())
+}
+
 pub fn get_audio_wav_u8(audio: &Tensor, sample_rate: u32) -> Result<Vec<u8>> {
     let spec = hound::WavSpec {
         channels: 1,

src/utils/mod.rs

@@ -727,7 +727,8 @@ pub fn contains_cjk(text: &str) -> bool {
         if (0x4e00..=0x9fff).contains(&c) // CJK Unified Ideographs
             || (0x3400..=0x4dbf).contains(&c) // CJK Unified Ideographs Extension A
             || (0x3040..=0x30ff).contains(&c) // Hiragana and Katakana
-            || (0xac00..=0xd7af).contains(&c) // Hangul Syllables
+            || (0xac00..=0xd7af).contains(&c)
+        // Hangul Syllables
         {
             return true;
         }
@@ -737,10 +738,10 @@ pub fn contains_cjk(text: &str) -> bool {
 
 pub fn prepare_tts_text(text: &str) -> Result<String> {
     let mut normalized_text = text.trim().to_string();
-    if normalized_text.eq("") {
+    if normalized_text.is_empty() {
-        return Err(anyhow!("Text cannot be empty."))
+        return Err(anyhow!("Text cannot be empty."));
     }
-    normalized_text = normalized_text.replace('\n', " ").replace('\r', " ");
+    normalized_text = normalized_text.replace(['\n', '\r'], " ");
     while normalized_text.contains("  ") {
         normalized_text = normalized_text.replace("  ", " ");
     }
@@ -753,22 +754,22 @@ pub fn prepare_tts_text(text: &str) -> Result<String> {
         return Ok(normalized_text);
     }
 
-    // Non-CJK (English/Western) logic        
+    // Non-CJK (English/Western) logic
     // Capitalize first letter if it's lowercase alphabetic
-    if let Some(first_char) = normalized_text.chars().next() {
+    if let Some(first_char) = normalized_text.chars().next()
-        if first_char.is_ascii_lowercase() {
+        && first_char.is_ascii_lowercase()
-            let mut chars = normalized_text.chars();
+    {
-            chars.next(); // consume first char
+        let mut chars = normalized_text.chars();
-            let rest: String = chars.collect();
+        chars.next(); // consume first char
-            normalized_text = format!("{}{}", first_char.to_ascii_uppercase(), rest);
+        let rest: String = chars.collect();
-        }
+        normalized_text = format!("{}{}", first_char.to_ascii_uppercase(), rest);
     }
 
     // Add period if ends with alphanumeric
-    if let Some(last_char) = normalized_text.chars().last() {
+    if let Some(last_char) = normalized_text.chars().last()
-        if last_char.is_alphanumeric() {
+        && last_char.is_alphanumeric()
-            normalized_text.push('.');
+    {
-        }
+        normalized_text.push('.');
     }
 
     // Add padding if less than 5 words

tests/config_tests.rs

@@ -1,5 +1,5 @@
 use aha::models::{
-    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
+    deepseek_ocr::config::DeepseekOCRConfig, hunyuan_ocr::config::HunYuanVLConfig, lfm2::config::Lfm2Config, lfm2vl::config::{Lfm2ProcessorConfig, Lfm2VLConfig}, minicpm4::config::MiniCPM4Config, moss_audio_tokenizer_nano::config::MossAudioTokenizerConfig, moss_tts_nano::config::MossTTSConfig, paddleocr_vl::config::PaddleOCRVLConfig, qwen2_5vl::config::Qwen2_5VLConfig, qwen3vl::config::Qwen3VLConfig, voxcpm::config::VoxCPMConfig
 };
 use anyhow::Result;
 
@@ -128,4 +128,4 @@ fn moss_tts_config() -> Result<()> {
     let config: MossTTSConfig = serde_json::from_slice(&std::fs::read(config_path)?)?;
     println!("{:?}", config);
     Ok(())
-}
+}

tests/test_moss_tts.rs

@@ -1,4 +1,4 @@
-use aha::models::moss::generate::MossTTSGenerate;
+use aha::models::moss_tts_nano::generate::MossTTSGenerate;
 use anyhow::Result;
 
 #[test]
@@ -10,11 +10,13 @@ fn moss_tts() -> Result<()> {
     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",
+        "你在干吗啊",
-        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,
+        Some("file://./assets/audio/jiangjiang.wav"),
+        None,
+        // Some("哈喽大家好，我是蒋蒋"),
+        // Some(aha::models::moss_tts_nano::model::MossTTSMode::Continuation),
+        None,
     )?;
     Ok(())
 }

tests/test_voxcpm.rs

@@ -6,7 +6,7 @@ use aha::{
         GenerateModel,
         voxcpm::{generate::VoxCPMGenerate, tokenizer::SingleChineseTokenizer},
     },
-    utils::audio_utils::{extract_and_save_audio_from_response, save_wav},
+    utils::audio_utils::{extract_and_save_audio_from_response, save_wav_mono},
 };
 use anyhow::{Ok, Result};
 
@@ -54,7 +54,7 @@ fn voxcpm_use_message_generate() -> Result<()> {
     }
     let i_duration = i_start.elapsed();
     println!("Time elapsed in generate is: {:?}", i_duration);
-    // save_wav(&generate, "voxcpm.wav", 16000)?;
+    // save_wav_mono(&generate, "voxcpm.wav", 16000)?;
     Ok(())
 }
 
@@ -105,7 +105,7 @@ fn voxcpm_generate() -> Result<()> {
 
     let i_duration = i_start.elapsed();
     println!("Time elapsed in generate is: {:?}", i_duration);
-    save_wav(&generate, "voxcpm.wav", 16000)?;
+    save_wav_mono(&generate, "voxcpm.wav", 16000)?;
     Ok(())
 }
 

tests/test_voxcpm1_5.rs

@@ -7,7 +7,7 @@ use aha::{
         GenerateModel,
         voxcpm::{generate::VoxCPMGenerate, tokenizer::SingleChineseTokenizer},
     },
-    utils::audio_utils::{extract_and_save_audio_from_response, save_wav},
+    utils::audio_utils::{extract_and_save_audio_from_response, save_wav_mono},
 };
 use anyhow::{Ok, Result};
 
@@ -106,7 +106,7 @@ fn voxcpm1_5_generate() -> Result<()> {
 
     let i_duration = i_start.elapsed();
     println!("Time elapsed in generate is: {:?}", i_duration);
-    save_wav(&generate, "voxcpm1_5.wav", 44100)?;
+    save_wav_mono(&generate, "voxcpm1_5.wav", 44100)?;
     Ok(())
 }
 
@@ -169,7 +169,7 @@ fn voxcpm_refact_generate() -> Result<()> {
     std::thread::sleep(std::time::Duration::from_secs(2));
     let i_duration = i_start.elapsed();
     println!("Time elapsed in generate is: {:?}", i_duration);
-    save_wav(
+    save_wav_mono(
         &generate,
         "voxcpm.wav",
         voxcpm_generate.sample_rate() as u32,

tests/weight_test.rs

@@ -444,11 +444,14 @@ 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 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(())
-}
+}