fix: ci-cd

Print debug messages to stderr instead of stdout

.github/workflows/ci-cd.yaml

@@ -0,0 +1,33 @@
+name: Build container
+env:
+  VERSION: 0.0.3
+  REGISTRY: https://harbor.bwgdi.com
+  REGISTRY_NAME: harbor.bwgdi.com
+  REGISTRY_PATH: library
+  DOCKER_NAME: voxcpmtts
+on:
+  push:
+    branches:
+      - main
+  workflow_dispatch:
+jobs:
+  build-docker:
+    runs-on: builder-ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ secrets.BWGDI_NAME }}
+          password: ${{ secrets.BWGDI_TOKEN }}
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+      - name: Build and push
+        uses: docker/build-push-action@v4
+        with:
+          context: .
+          file: ./Dockerfile
+          push: true
+          tags: ${{ env.REGISTRY_NAME }}/${{ env.REGISTRY_PATH }}/${{ env.DOCKER_NAME }}:${{ env.VERSION }}

.gitignore

@@ -1,4 +1,42 @@
 launch.json
-__pycache__
 voxcpm.egg-info
 .DS_Store
+
+# Python-generated files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# Distribution / packaging
+build/
+dist/
+wheels/
+*.egg-info/
+
+# Unit test / coverage reports
+.pytest_cache/
+.coverage
+htmlcov/
+coverage.xml
+
+# Logs
+*.log
+log/*.log
+
+# Virtual environments
+.venv/
+venv/
+env/
+
+# IDE settings
+.vscode/
+.idea/
+
+# OS generated files
+.DS_Store
+
+# Generated files
+*.wav
+*.pdf
+
+*.lock

Dockerfile

@@ -0,0 +1,17 @@
+FROM python:3.10.12-slim
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    ffmpeg \
+    && rm -rf /var/lib/apt/lists/*
+# Create app directory
+WORKDIR /app
+COPY api_concurrent.py requirements.txt ./
+RUN pip install -r requirements.txt
+ENV VOXCPM_MODEL_ID="/models/VoxCPM1.5/" \
+    VOXCPM_CPU_WORKERS="2" \
+    VOXCPM_UVICORN_WORKERS="1" \
+    MAX_GPU_CONCURRENT="1"
+
+EXPOSE 5000
+CMD [ "python", "./api_concurrent.py" ]
+

README_BW.md

@@ -0,0 +1,83 @@
+# VoxCPM-TTS
+
+https://github.com/BoardWare-Genius/VoxCPM
+
+## 📦 VoxCPM-TTS Version History
+
+| Version | Date       | Summary                         |
+|---------|------------|---------------------------------|
+| 0.0.3   | 2026-02-09 | Optimized configuration & Model support |
+| 0.0.2   | 2026-01-21 | Supports streaming               |
+| 0.0.1   | 2026-01-20 | Initial version                 |
+
+### 🔄 Version Details
+
+#### 🆕 0.0.3 – *2026-02-09*
+- ✅ **Configuration & Deployment**
+  - Supports configuring model path via `VOXCPM_MODEL_ID`
+  - Supports configuring CPU workers via `VOXCPM_CPU_WORKERS`
+  - Supports configuring Uvicorn workers via `VOXCPM_UVICORN_WORKERS`
+
+#### 🆕 0.0.2 – *2026-01-21*
+
+- ✅ **Core Features**
+  - Update Model Weights, use VoxCPM1.5 and model parameters
+  - Supports streaming 
+
+#### 🆕 0.0.1 – *2026-01-20*
+
+- ✅ **Core Features**
+  - Initial VoxCPM-TTS
+
+---
+
+
+# Start
+```bash
+docker pull harbor.bwgdi.com/library/voxcpmtts:0.0.3
+
+# Run with custom configuration
+# -e VOXCPM_MODEL_ID: Path to the model directory inside container
+# -e VOXCPM_CPU_WORKERS: Number of threads for CPU-bound tasks
+# -e VOXCPM_UVICORN_WORKERS: Number of uvicorn workers
+# -e MAX_GPU_CONCURRENT: Max concurrent GPU tasks
+docker run -d --restart always -p 5001:5000 --gpus '"device=0"' \
+  -e VOXCPM_MODEL_ID="/models/VoxCPM1.5/" \
+  -e VOXCPM_CPU_WORKERS="2" \
+  -e VOXCPM_UVICORN_WORKERS="1" \
+  -e MAX_GPU_CONCURRENT="1" \
+  --mount type=bind,source=/Workspace/NAS11/model/Voice/VoxCPM,target=/models \
+  harbor.bwgdi.com/library/voxcpmtts:0.0.3
+```
+
+# Usage
+
+## Non-streaming
+```bash
+curl --location 'http://localhost:5001/generate_tts' \
+--form 'text="你好，这是一段测试文本"' \
+--form 'prompt_text="这是提示文本"' \
+--form 'cfg_value="2.0"' \
+--form 'inference_timesteps="10"' \
+--form 'do_normalize="true"' \
+--form 'denoise="true"' \
+--form 'retry_badcase="true"' \
+--form 'retry_badcase_max_times="3"' \
+--form 'retry_badcase_ratio_threshold="6.0"' \
+--form 'prompt_wav=@"/assets/2food16k_2.wav"'
+```
+
+## Streaming
+```bash
+curl --location 'http://localhost:5001/generate_tts_streaming' \
+--form 'text="你好，这是一段测试文本"' \
+--form 'prompt_text="这是提示文本"' \
+--form 'cfg_value="2.0"' \
+--form 'inference_timesteps="10"' \
+--form 'do_normalize="true"' \
+--form 'denoise="true"' \
+--form 'retry_badcase="true"' \
+--form 'retry_badcase_max_times="3"' \
+--form 'retry_badcase_ratio_threshold="6.0"' \
+--form 'prompt_wav=@"/Workspace/NAS11/model/Voice/assets/2food16k_2.wav"'
+```

api_concurrent.py

@@ -0,0 +1,272 @@
+import os
+import torch
+import tempfile
+import soundfile as sf
+import numpy as np
+import wave
+from io import BytesIO
+import asyncio
+from typing import Optional
+from fastapi import FastAPI, Form, UploadFile, BackgroundTasks
+from fastapi.responses import StreamingResponse, JSONResponse
+from concurrent.futures import ThreadPoolExecutor
+
+import voxcpm
+import uvicorn
+
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+if os.environ.get("HF_REPO_ID", "").strip() == "":
+    os.environ["HF_REPO_ID"] = "/models/VoxCPM1.5/"
+
+
+# ========== 模型类 ==========
+class VoxCPMDemo:
+    def __init__(self) -> None:
+        self.device = "cuda" if torch.cuda.is_available() else "cpu"
+        print(f"🚀 Running on device: {self.device}")
+        self.voxcpm_model: Optional[voxcpm.VoxCPM] = None
+        self.default_local_model_dir = os.environ.get("VOXCPM_MODEL_ID", "/models/VoxCPM1.5/")
+
+    def _resolve_model_dir(self) -> str:
+        if os.path.isdir(self.default_local_model_dir):
+            return self.default_local_model_dir
+        return "models"
+
+    def get_or_load_voxcpm(self) -> voxcpm.VoxCPM:
+        if self.voxcpm_model is None:
+            print("🔄 Loading VoxCPM model...")
+            model_dir = self._resolve_model_dir()
+            self.voxcpm_model = voxcpm.VoxCPM(voxcpm_model_path=model_dir)
+            print("✅ VoxCPM model loaded.")
+        return self.voxcpm_model
+
+    def tts_generate(
+        self,
+        text: str,
+        prompt_wav_path: Optional[str] = None,
+        prompt_text: Optional[str] = None,
+        cfg_value: float = 2.0,
+        inference_timesteps: int = 10,
+        do_normalize: bool = True,
+        denoise: bool = True,
+        retry_badcase: bool = True,
+        retry_badcase_max_times: int = 3,
+        retry_badcase_ratio_threshold: float = 6.0,
+    ) -> str:
+        model = self.get_or_load_voxcpm()
+        wav = model.generate(
+            text=text,
+            prompt_text=prompt_text,
+            prompt_wav_path=prompt_wav_path,
+            cfg_value=float(cfg_value),
+            inference_timesteps=int(inference_timesteps),
+            normalize=do_normalize,
+            denoise=denoise,
+            retry_badcase=retry_badcase,
+            retry_badcase_max_times=retry_badcase_max_times,
+            retry_badcase_ratio_threshold=retry_badcase_ratio_threshold,
+        )
+        tmp_wav = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
+        sf.write(tmp_wav.name, wav, model.tts_model.sample_rate)
+        torch.cuda.empty_cache()
+        return tmp_wav.name
+
+    def tts_generate_streaming(
+        self,
+        text: str,
+        prompt_wav_path: Optional[str] = None,
+        prompt_text: Optional[str] = None,
+        cfg_value: float = 2.0,
+        inference_timesteps: int = 10,
+        do_normalize: bool = True,
+        denoise: bool = True,
+        retry_badcase: bool = True,
+        retry_badcase_max_times: int = 3,
+        retry_badcase_ratio_threshold: float = 6.0,
+    ):
+        """Generates audio and yields it as a stream of WAV chunks."""
+        model = self.get_or_load_voxcpm()
+
+        # 1. Yield a WAV header first.
+        # The size fields will be 0, which is standard for streaming.
+        SAMPLE_RATE = model.tts_model.sample_rate
+        CHANNELS = 1
+        SAMPLE_WIDTH = 2  # 16-bit
+
+        header_buf = BytesIO()
+        with wave.open(header_buf, "wb") as wf:
+            wf.setnchannels(CHANNELS)
+            wf.setsampwidth(SAMPLE_WIDTH)
+            wf.setframerate(SAMPLE_RATE)
+
+        yield header_buf.getvalue()
+
+        # 2. Generate and yield audio chunks.
+        # NOTE: We assume a `generate_stream` method exists on the model that yields audio chunks.
+        # You may need to change `generate_stream` to the actual method name in your version of voxcpm.
+        try:
+            stream = model.generate_streaming(
+                text=text,
+                prompt_text=prompt_text,
+                prompt_wav_path=prompt_wav_path,
+                cfg_value=float(cfg_value),
+                inference_timesteps=int(inference_timesteps),
+                normalize=do_normalize,
+                denoise=denoise,
+                retry_badcase=retry_badcase,
+                retry_badcase_max_times=retry_badcase_max_times,
+                retry_badcase_ratio_threshold=retry_badcase_ratio_threshold,
+            )
+            for chunk_np in stream:  # Assuming it yields numpy arrays
+                # Ensure audio is in 16-bit PCM format for streaming
+                if chunk_np.dtype in [np.float32, np.float64]:
+                    chunk_np = (chunk_np * 32767).astype(np.int16)
+                yield chunk_np.tobytes()
+        finally:
+            torch.cuda.empty_cache()
+
+
+# ========== FastAPI ==========
+app = FastAPI(title="VoxCPM API", version="1.0.0")
+demo = VoxCPMDemo()
+
+# --- Concurrency Control ---
+# Use a semaphore to limit concurrent GPU tasks.
+MAX_GPU_CONCURRENT = int(os.environ.get("MAX_GPU_CONCURRENT", "1"))
+gpu_semaphore = asyncio.Semaphore(MAX_GPU_CONCURRENT)
+
+# Use a thread pool for running blocking (CPU/GPU-bound) code.
+MAX_CPU_WORKERS = int(os.environ.get("VOXCPM_CPU_WORKERS", "2"))
+executor = ThreadPoolExecutor(max_workers=MAX_CPU_WORKERS)
+
+@app.on_event("shutdown")
+def shutdown_event():
+    print("Shutting down thread pool executor...")
+    executor.shutdown(wait=True)
+
+
+# ---------- TTS API ----------
+@app.post("/generate_tts")
+async def generate_tts(
+    background_tasks: BackgroundTasks,
+    text: str = Form(...),
+    prompt_text: Optional[str] = Form(None),
+    cfg_value: float = Form(2.0),
+    inference_timesteps: int = Form(10),
+    do_normalize: bool = Form(True),
+    denoise: bool = Form(True),
+    retry_badcase: bool = Form(True),
+    retry_badcase_max_times: int = Form(3),
+    retry_badcase_ratio_threshold: float = Form(6.0),
+    prompt_wav: Optional[UploadFile] = None,
+):
+
+    try:
+        prompt_path = None
+        if prompt_wav:
+            with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
+                tmp.write(await prompt_wav.read())
+                prompt_path = tmp.name
+                background_tasks.add_task(os.remove, tmp.name)
+
+        # Submit to GPU via semaphore and executor
+        loop = asyncio.get_running_loop()
+        async with gpu_semaphore:
+            output_path = await loop.run_in_executor(
+                executor,
+                demo.tts_generate,
+                text,
+                prompt_path,
+                prompt_text,
+                cfg_value,
+                inference_timesteps,
+                do_normalize,
+                denoise,
+                retry_badcase,
+                retry_badcase_max_times,
+                retry_badcase_ratio_threshold,
+            )
+
+        # 后台删除生成的文件
+        background_tasks.add_task(os.remove, output_path)
+
+        return StreamingResponse(
+            open(output_path, "rb"),
+            media_type="audio/wav",
+            headers={"Content-Disposition": 'attachment; filename="output.wav"'}
+        )
+
+    except Exception as e:
+        return JSONResponse({"error": str(e)}, status_code=500)
+
+
+@app.post("/generate_tts_streaming")
+async def generate_tts_streaming(
+    background_tasks: BackgroundTasks,
+    text: str = Form(...),
+    prompt_text: Optional[str] = Form(None),
+    cfg_value: float = Form(2.0),
+    inference_timesteps: int = Form(10),
+    do_normalize: bool = Form(True),
+    denoise: bool = Form(True),
+    retry_badcase: bool = Form(True),
+    retry_badcase_max_times: int = Form(3),
+    retry_badcase_ratio_threshold: float = Form(6.0),
+    prompt_wav: Optional[UploadFile] = None,
+):
+    try:
+        prompt_path = None
+        if prompt_wav:
+            # Save uploaded file to a temporary file
+            with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
+                tmp.write(await prompt_wav.read())
+                prompt_path = tmp.name
+                # Ensure the temp file is deleted after the request is finished
+                background_tasks.add_task(os.remove, prompt_path)
+
+        async def stream_generator():
+            # This async generator consumes from a queue populated by a sync generator in a thread.
+            q = asyncio.Queue()
+            loop = asyncio.get_running_loop()
+
+            def producer():
+                # This runs in the executor thread and produces chunks.
+                try:
+                    # This is a sync generator
+                    for chunk in demo.tts_generate_streaming(
+                        text, prompt_path, prompt_text, cfg_value,
+                        inference_timesteps, do_normalize, denoise,
+                        retry_badcase, retry_badcase_max_times, retry_badcase_ratio_threshold
+                    ):
+                        loop.call_soon_threadsafe(q.put_nowait, chunk)
+                except Exception as e:
+                    # Put the exception in the queue to be re-raised in the consumer
+                    loop.call_soon_threadsafe(q.put_nowait, e)
+                finally:
+                    # Signal the end of the stream
+                    loop.call_soon_threadsafe(q.put_nowait, None)
+            
+            # Acquire the GPU semaphore before starting the producer thread.
+            async with gpu_semaphore:
+                loop.run_in_executor(executor, producer)
+                while True:
+                    chunk = await q.get()
+                    if chunk is None:
+                        break
+                    if isinstance(chunk, Exception):
+                        raise chunk
+                    yield chunk
+        
+        return StreamingResponse(stream_generator(), media_type="audio/wav")
+
+    except Exception as e:
+        return JSONResponse({"error": str(e)}, status_code=500)
+
+
+@app.get("/")
+async def root():
+    return {"message": "VoxCPM API running 🚀", "endpoints": ["/generate_tts"]}
+
+if __name__ == "__main__":
+    uvicorn_workers = int(os.environ.get("VOXCPM_UVICORN_WORKERS", "1"))
+    uvicorn.run("api_concurrent:app", host="0.0.0.0", port=5000, workers=uvicorn_workers)

app.py

@@ -1,4 +1,5 @@
 import os
+import sys
 import numpy as np
 import torch
 import gradio as gr  
@@ -16,7 +17,7 @@ import voxcpm
 class VoxCPMDemo:
     def __init__(self) -> None:
         self.device = "cuda" if torch.cuda.is_available() else "cpu"
-        print(f"🚀 Running on device: {self.device}")
+        print(f"🚀 Running on device: {self.device}", file=sys.stderr)
 
         # ASR model for prompt text recognition
         self.asr_model_id = "iic/SenseVoiceSmall"
@@ -49,10 +50,10 @@ class VoxCPMDemo:
                 try:
                     from huggingface_hub import snapshot_download  # type: ignore
                     os.makedirs(target_dir, exist_ok=True)
-                    print(f"Downloading model from HF repo '{repo_id}' to '{target_dir}' ...")
+                    print(f"Downloading model from HF repo '{repo_id}' to '{target_dir}' ...", file=sys.stderr)
                     snapshot_download(repo_id=repo_id, local_dir=target_dir, local_dir_use_symlinks=False)
                 except Exception as e:
-                    print(f"Warning: HF download failed: {e}. Falling back to 'data'.")
+                    print(f"Warning: HF download failed: {e}. Falling back to 'data'.", file=sys.stderr)
                     return "models"
             return target_dir
         return "models"
@@ -60,11 +61,11 @@ class VoxCPMDemo:
     def get_or_load_voxcpm(self) -> voxcpm.VoxCPM:
         if self.voxcpm_model is not None:
             return self.voxcpm_model
-        print("Model not loaded, initializing...")
+        print("Model not loaded, initializing...", file=sys.stderr)
         model_dir = self._resolve_model_dir()
-        print(f"Using model dir: {model_dir}")
+        print(f"Using model dir: {model_dir}", file=sys.stderr)
         self.voxcpm_model = voxcpm.VoxCPM(voxcpm_model_path=model_dir)
-        print("Model loaded successfully.")
+        print("Model loaded successfully.", file=sys.stderr)
         return self.voxcpm_model
 
     # ---------- Functional endpoints ----------
@@ -98,7 +99,7 @@ class VoxCPMDemo:
         prompt_wav_path = prompt_wav_path_input if prompt_wav_path_input else None
         prompt_text = prompt_text_input if prompt_text_input else None
 
-        print(f"Generating audio for text: '{text[:60]}...'")
+        print(f"Generating audio for text: '{text[:60]}...'", file=sys.stderr)
         wav = current_model.generate(
             text=text,
             prompt_text=prompt_text,

lora_ft_webui.py

@@ -104,7 +104,7 @@ def get_timestamp_str():
 def get_or_load_asr_model():
     global asr_model
     if asr_model is None:
-        print("Loading ASR model (SenseVoiceSmall)...")
+        print("Loading ASR model (SenseVoiceSmall)...", file=sys.stderr)
         device = "cuda:0" if torch.cuda.is_available() else "cpu"
         asr_model = AutoModel(
             model="iic/SenseVoiceSmall",
@@ -123,7 +123,7 @@ def recognize_audio(audio_path):
         text = res[0]["text"].split('|>')[-1]
         return text
     except Exception as e:
-        print(f"ASR Error: {e}")
+        print(f"ASR Error: {e}", file=sys.stderr)
         return ""
 
 def scan_lora_checkpoints(root_dir="lora", with_info=False):
@@ -181,7 +181,7 @@ def load_lora_config_from_checkpoint(lora_path):
             if lora_cfg_dict:
                 return LoRAConfig(**lora_cfg_dict), lora_info.get("base_model")
         except Exception as e:
-            print(f"Warning: Failed to load lora_config.json: {e}")
+            print(f"Warning: Failed to load lora_config.json: {e}", file=sys.stderr)
     return None, None
 
 def get_default_lora_config():
@@ -197,7 +197,7 @@ def get_default_lora_config():
 
 def load_model(pretrained_path, lora_path=None):
     global current_model
-    print(f"Loading model from {pretrained_path}...")
+    print(f"Loading model from {pretrained_path}...", file=sys.stderr)
     
     lora_config = None
     lora_weights_path = None
@@ -209,11 +209,11 @@ def load_model(pretrained_path, lora_path=None):
             # Try to load LoRA config from lora_config.json
             lora_config, _ = load_lora_config_from_checkpoint(full_lora_path)
             if lora_config:
-                print(f"Loaded LoRA config from {full_lora_path}/lora_config.json")
+                print(f"Loaded LoRA config from {full_lora_path}/lora_config.json", file=sys.stderr)
             else:
                 # Fallback to default config for old checkpoints
                 lora_config = get_default_lora_config()
-                print("Using default LoRA config (lora_config.json not found)")
+                print("Using default LoRA config (lora_config.json not found)", file=sys.stderr)
     
     # Always init with a default LoRA config to allow hot-swapping later
     if lora_config is None:
@@ -251,36 +251,36 @@ def run_inference(text, prompt_wav, prompt_text, lora_selection, cfg_scale, step
                         # 优先使用保存的 base_model 路径
                         if os.path.exists(saved_base_model):
                             base_model_path = saved_base_model
-                            print(f"Using base model from LoRA config: {base_model_path}")
+                            print(f"Using base model from LoRA config: {base_model_path}", file=sys.stderr)
                         else:
-                            print(f"Warning: Saved base_model path not found: {saved_base_model}")
-                            print(f"Falling back to default: {base_model_path}")
+                            print(f"Warning: Saved base_model path not found: {saved_base_model}", file=sys.stderr)
+                            print(f"Falling back to default: {base_model_path}", file=sys.stderr)
                 except Exception as e:
-                    print(f"Warning: Failed to read base_model from LoRA config: {e}")
+                    print(f"Warning: Failed to read base_model from LoRA config: {e}", file=sys.stderr)
         
         # 加载模型
         try:
-            print(f"Loading base model: {base_model_path}")
+            print(f"Loading base model: {base_model_path}", file=sys.stderr)
             status_msg = load_model(base_model_path)
             if lora_selection and lora_selection != "None":
-                print(f"Model loaded for LoRA: {lora_selection}")
+                print(f"Model loaded for LoRA: {lora_selection}", file=sys.stderr)
         except Exception as e:
             error_msg = f"Failed to load model from {base_model_path}: {str(e)}"
-            print(error_msg)
+            print(error_msg, file=sys.stderr)
             return None, error_msg
 
     # Handle LoRA hot-swapping
     if lora_selection and lora_selection != "None":
         full_lora_path = os.path.join("lora", lora_selection)
-        print(f"Hot-loading LoRA: {full_lora_path}")
+        print(f"Hot-loading LoRA: {full_lora_path}", file=sys.stderr)
         try:
             current_model.load_lora(full_lora_path)
             current_model.set_lora_enabled(True)
         except Exception as e:
-            print(f"Error loading LoRA: {e}")
+            print(f"Error loading LoRA: {e}", file=sys.stderr)
             return None, f"Error loading LoRA: {e}"
     else:
-        print("Disabling LoRA")
+        print("Disabling LoRA", file=sys.stderr)
         current_model.set_lora_enabled(False)
 
     if seed != -1:
@@ -297,11 +297,11 @@ def run_inference(text, prompt_wav, prompt_text, lora_selection, cfg_scale, step
         
         # 如果没有提供参考文本，尝试自动识别
         if not prompt_text or not prompt_text.strip():
-            print("参考音频已提供但缺少文本，自动识别中...")
+            print("参考音频已提供但缺少文本，自动识别中...", file=sys.stderr)
             try:
                 final_prompt_text = recognize_audio(prompt_wav)
                 if final_prompt_text:
-                    print(f"自动识别文本: {final_prompt_text}")
+                    print(f"自动识别文本: {final_prompt_text}", file=sys.stderr)
                 else:
                     return None, "错误：无法识别参考音频内容，请手动填写参考文本"
             except Exception as e:
@@ -1114,12 +1114,12 @@ with gr.Blocks(
                 choices = ["None"] + [ckpt[0] for ckpt in checkpoints_with_info]
                 
                 # 输出调试信息
-                print(f"刷新 LoRA 列表: 找到 {len(checkpoints_with_info)} 个检查点")
+                print(f"刷新 LoRA 列表: 找到 {len(checkpoints_with_info)} 个检查点", file=sys.stderr)
                 for ckpt_path, base_model in checkpoints_with_info:
                     if base_model:
-                        print(f"  - {ckpt_path} (Base Model: {base_model})")
+                        print(f"  - {ckpt_path} (Base Model: {base_model})", file=sys.stderr)
                     else:
-                        print(f"  - {ckpt_path}")
+                        print(f"  - {ckpt_path}", file=sys.stderr)
                 
                 return gr.update(choices=choices, value="None")
 

pyproject.toml

@@ -27,6 +27,7 @@ requires-python = ">=3.10"
 dependencies = [
     "torch>=2.5.0",
     "torchaudio>=2.5.0",
+    "torchcodec",
     "transformers>=4.36.2",
     "einops",
     "gradio<6",
@@ -46,8 +47,7 @@ dependencies = [
     "funasr",
     "spaces",
     "argbind",
-    "safetensors"
-
+    "safetensors",
 ]
 
 [project.optional-dependencies]

requirements.txt

@@ -0,0 +1,5 @@
+soundfile
+fastapi
+uvicorn
+voxcpm
+torchcodec

scripts/test_voxcpm_ft_infer.py

@@ -23,6 +23,7 @@ With voice cloning:
 """
 
 import argparse
+import sys
 from pathlib import Path
 
 import soundfile as sf
@@ -92,7 +93,7 @@ def main():
     args = parse_args()
 
     # Load model from checkpoint directory (no denoiser)
-    print(f"[FT Inference] Loading model: {args.ckpt_dir}")
+    print(f"[FT Inference] Loading model: {args.ckpt_dir}", file=sys.stderr)
     model = VoxCPM.from_pretrained(
         hf_model_id=args.ckpt_dir,
         load_denoiser=False,
@@ -103,10 +104,10 @@ def main():
     prompt_wav_path = args.prompt_audio if args.prompt_audio else None
     prompt_text = args.prompt_text if args.prompt_text else None
 
-    print(f"[FT Inference] Synthesizing: text='{args.text}'")
+    print(f"[FT Inference] Synthesizing: text='{args.text}'", file=sys.stderr)
     if prompt_wav_path:
-        print(f"[FT Inference] Using reference audio: {prompt_wav_path}")
-        print(f"[FT Inference] Reference text: {prompt_text}")
+        print(f"[FT Inference] Using reference audio: {prompt_wav_path}", file=sys.stderr)
+        print(f"[FT Inference] Reference text: {prompt_text}", file=sys.stderr)
 
     audio_np = model.generate(
         text=args.text,
@@ -124,7 +125,7 @@ def main():
     out_path.parent.mkdir(parents=True, exist_ok=True)
     sf.write(str(out_path), audio_np, model.tts_model.sample_rate)
 
-    print(f"[FT Inference] Saved to: {out_path}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s")
+    print(f"[FT Inference] Saved to: {out_path}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s", file=sys.stderr)
 
 
 if __name__ == "__main__":

scripts/test_voxcpm_lora_infer.py

@@ -24,6 +24,7 @@ Note: The script reads base_model path and lora_config from lora_config.json
 
 import argparse
 import json
+import sys
 from pathlib import Path
 
 import soundfile as sf
@@ -124,13 +125,13 @@ def main():
     lora_cfg_dict = lora_info.get("lora_config", {})
     lora_cfg = LoRAConfig(**lora_cfg_dict) if lora_cfg_dict else None
     
-    print(f"Loaded config from: {lora_config_path}")
-    print(f"  Base model: {pretrained_path}")
-    print(f"  LoRA config: r={lora_cfg.r}, alpha={lora_cfg.alpha}" if lora_cfg else "  LoRA config: None")
+    print(f"Loaded config from: {lora_config_path}", file=sys.stderr)
+    print(f"  Base model: {pretrained_path}", file=sys.stderr)
+    print(f"  LoRA config: r={lora_cfg.r}, alpha={lora_cfg.alpha}" if lora_cfg else "  LoRA config: None", file=sys.stderr)
 
     # 3. Load model with LoRA (no denoiser)
-    print(f"\n[1/2] Loading model with LoRA: {pretrained_path}")
-    print(f"      LoRA weights: {ckpt_dir}")
+    print(f"\n[1/2] Loading model with LoRA: {pretrained_path}", file=sys.stderr)
+    print(f"      LoRA weights: {ckpt_dir}", file=sys.stderr)
     model = VoxCPM.from_pretrained(
         hf_model_id=pretrained_path,
         load_denoiser=False,
@@ -145,10 +146,10 @@ def main():
     out_path = Path(args.output)
     out_path.parent.mkdir(parents=True, exist_ok=True)
 
-    print(f"\n[2/2] Starting synthesis tests...")
+    print(f"\n[2/2] Starting synthesis tests...", file=sys.stderr)
     
     # === Test 1: With LoRA ===
-    print(f"\n  [Test 1] Synthesize with LoRA...")
+    print(f"\n  [Test 1] Synthesize with LoRA...", file=sys.stderr)
     audio_np = model.generate(
         text=args.text,
         prompt_wav_path=prompt_wav_path,
@@ -161,10 +162,10 @@ def main():
     )
     lora_output = out_path.with_stem(out_path.stem + "_with_lora")
     sf.write(str(lora_output), audio_np, model.tts_model.sample_rate)
-    print(f"           Saved: {lora_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s")
+    print(f"           Saved: {lora_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s", file=sys.stderr)
 
     # === Test 2: Disable LoRA (via set_lora_enabled) ===
-    print(f"\n  [Test 2] Disable LoRA (set_lora_enabled=False)...")
+    print(f"\n  [Test 2] Disable LoRA (set_lora_enabled=False)...", file=sys.stderr)
     model.set_lora_enabled(False)
     audio_np = model.generate(
         text=args.text,
@@ -178,10 +179,10 @@ def main():
     )
     disabled_output = out_path.with_stem(out_path.stem + "_lora_disabled")
     sf.write(str(disabled_output), audio_np, model.tts_model.sample_rate)
-    print(f"           Saved: {disabled_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s")
+    print(f"           Saved: {disabled_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s", file=sys.stderr)
 
     # === Test 3: Re-enable LoRA ===
-    print(f"\n  [Test 3] Re-enable LoRA (set_lora_enabled=True)...")
+    print(f"\n  [Test 3] Re-enable LoRA (set_lora_enabled=True)...", file=sys.stderr)
     model.set_lora_enabled(True)
     audio_np = model.generate(
         text=args.text,
@@ -195,10 +196,10 @@ def main():
     )
     reenabled_output = out_path.with_stem(out_path.stem + "_lora_reenabled")
     sf.write(str(reenabled_output), audio_np, model.tts_model.sample_rate)
-    print(f"           Saved: {reenabled_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s")
+    print(f"           Saved: {reenabled_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s", file=sys.stderr)
 
     # === Test 4: Unload LoRA (reset_lora_weights) ===
-    print(f"\n  [Test 4] Unload LoRA (unload_lora)...")
+    print(f"\n  [Test 4] Unload LoRA (unload_lora)...", file=sys.stderr)
     model.unload_lora()
     audio_np = model.generate(
         text=args.text,
@@ -212,12 +213,12 @@ def main():
     )
     reset_output = out_path.with_stem(out_path.stem + "_lora_reset")
     sf.write(str(reset_output), audio_np, model.tts_model.sample_rate)
-    print(f"           Saved: {reset_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s")
+    print(f"           Saved: {reset_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s", file=sys.stderr)
 
     # === Test 5: Hot-reload LoRA (load_lora) ===
-    print(f"\n  [Test 5] Hot-reload LoRA (load_lora)...")
+    print(f"\n  [Test 5] Hot-reload LoRA (load_lora)...", file=sys.stderr)
     loaded, skipped = model.load_lora(ckpt_dir)
-    print(f"           Reloaded {len(loaded)} parameters")
+    print(f"           Reloaded {len(loaded)} parameters", file=sys.stderr)
     audio_np = model.generate(
         text=args.text,
         prompt_wav_path=prompt_wav_path,
@@ -230,14 +231,14 @@ def main():
     )
     reload_output = out_path.with_stem(out_path.stem + "_lora_reloaded")
     sf.write(str(reload_output), audio_np, model.tts_model.sample_rate)
-    print(f"           Saved: {reload_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s")
+    print(f"           Saved: {reload_output}, duration: {len(audio_np) / model.tts_model.sample_rate:.2f}s", file=sys.stderr)
 
-    print(f"\n[Done] All tests completed!")
-    print(f"  - with_lora:      {lora_output}")
-    print(f"  - lora_disabled:  {disabled_output}")
-    print(f"  - lora_reenabled: {reenabled_output}")
-    print(f"  - lora_reset:     {reset_output}")
-    print(f"  - lora_reloaded:  {reload_output}")
+    print(f"\n[Done] All tests completed!", file=sys.stderr)
+    print(f"  - with_lora:      {lora_output}", file=sys.stderr)
+    print(f"  - lora_disabled:  {disabled_output}", file=sys.stderr)
+    print(f"  - lora_reenabled: {reenabled_output}", file=sys.stderr)
+    print(f"  - lora_reset:     {reset_output}", file=sys.stderr)
+    print(f"  - lora_reloaded:  {reload_output}", file=sys.stderr)
 
 
 if __name__ == "__main__":

scripts/train_voxcpm_finetune.py

@@ -24,7 +24,7 @@ try:
     SAFETENSORS_AVAILABLE = True
 except ImportError:
     SAFETENSORS_AVAILABLE = False
-    print("Warning: safetensors not available, will use pytorch format")
+    print("Warning: safetensors not available, will use pytorch format", file=sys.stderr)
 
 from voxcpm.model import VoxCPMModel
 from voxcpm.model.voxcpm import LoRAConfig
@@ -170,7 +170,7 @@ def train(
     # Only print param info on rank 0 to avoid cluttered output
     if accelerator.rank == 0:
         for name, param in model.named_parameters():
-            print(name, param.requires_grad)
+            print(name, param.requires_grad, file=sys.stderr)
 
     optimizer = AdamW(
         (p for p in model.parameters() if p.requires_grad),
@@ -210,12 +210,12 @@ def train(
             cur_step = int(_resume.get("step", start_step))
         except Exception:
             cur_step = start_step
-        print(f"Signal {signum} received. Saving checkpoint at step {cur_step} ...")
+        print(f"Signal {signum} received. Saving checkpoint at step {cur_step} ...", file=sys.stderr)
         try:
             save_checkpoint(_model, _optim, _sched, _save_dir, cur_step, _pretrained, _hf_id, _dist)
-            print("Checkpoint saved. Exiting.")
+            print("Checkpoint saved. Exiting.", file=sys.stderr)
         except Exception as e:
-            print(f"Error saving checkpoint on signal: {e}")
+            print(f"Error saving checkpoint on signal: {e}", file=sys.stderr)
         os._exit(0)
 
     signal.signal(signal.SIGTERM, _signal_handler)
@@ -553,7 +553,7 @@ def load_checkpoint(model, optimizer, scheduler, save_dir: Path):
             
             # Load only lora weights
             unwrapped.load_state_dict(state_dict, strict=False)
-            print(f"Loaded LoRA weights from {lora_weights_path}")
+            print(f"Loaded LoRA weights from {lora_weights_path}", file=sys.stderr)
     else:
         # Full finetune: load model.safetensors or pytorch_model.bin
         model_path = latest_folder / "model.safetensors"
@@ -569,26 +569,26 @@ def load_checkpoint(model, optimizer, scheduler, save_dir: Path):
                 state_dict = ckpt.get("state_dict", ckpt)
             
             unwrapped.load_state_dict(state_dict, strict=False)
-            print(f"Loaded model weights from {model_path}")
+            print(f"Loaded model weights from {model_path}", file=sys.stderr)
     
     # Load optimizer state
     optimizer_path = latest_folder / "optimizer.pth"
     if optimizer_path.exists():
         optimizer.load_state_dict(torch.load(optimizer_path, map_location="cpu"))
-        print(f"Loaded optimizer state from {optimizer_path}")
+        print(f"Loaded optimizer state from {optimizer_path}", file=sys.stderr)
     
     # Load scheduler state
     scheduler_path = latest_folder / "scheduler.pth"
     if scheduler_path.exists():
         scheduler.load_state_dict(torch.load(scheduler_path, map_location="cpu"))
-        print(f"Loaded scheduler state from {scheduler_path}")
+        print(f"Loaded scheduler state from {scheduler_path}", file=sys.stderr)
     
     # Try to infer step from checkpoint folders
     step_folders = [d for d in save_dir.iterdir() if d.is_dir() and d.name.startswith("step_")]
     if step_folders:
         steps = [int(d.name.split("_")[1]) for d in step_folders]
         resume_step = max(steps)
-        print(f"Resuming from step {resume_step}")
+        print(f"Resuming from step {resume_step}", file=sys.stderr)
         return resume_step
     
     return 0
@@ -670,7 +670,7 @@ def save_checkpoint(model, optimizer, scheduler, save_dir: Path, step: int, pret
                     latest_link.unlink()
             shutil.copytree(folder, latest_link)
         except Exception:
-            print(f"Warning: failed to update latest checkpoint link at {latest_link}")
+            print(f"Warning: failed to update latest checkpoint link at {latest_link}", file=sys.stderr)
 
 
 if __name__ == "__main__":

src/voxcpm/cli.py

@@ -45,7 +45,7 @@ def load_model(args) -> VoxCPM:
 
     Prefer --model-path if provided; otherwise use from_pretrained (Hub).
     """
-    print("Loading VoxCPM model...")
+    print("Loading VoxCPM model...", file=sys.stderr)
 
     # 兼容旧参数：ZIPENHANCER_MODEL_PATH 环境变量作为默认
     zipenhancer_path = getattr(args, "zipenhancer_path", None) or os.environ.get(
@@ -66,7 +66,7 @@ def load_model(args) -> VoxCPM:
             dropout=getattr(args, "lora_dropout", 0.0),
         )
         print(f"LoRA config: r={lora_config.r}, alpha={lora_config.alpha}, "
-              f"lm={lora_config.enable_lm}, dit={lora_config.enable_dit}, proj={lora_config.enable_proj}")
+              f"lm={lora_config.enable_lm}, dit={lora_config.enable_dit}, proj={lora_config.enable_proj}", file=sys.stderr)
 
     # Load from local path if provided
     if getattr(args, "model_path", None):
@@ -78,10 +78,10 @@ def load_model(args) -> VoxCPM:
                 lora_config=lora_config,
                 lora_weights_path=lora_weights_path,
             )
-            print("Model loaded (local).")
+            print("Model loaded (local).", file=sys.stderr)
             return model
         except Exception as e:
-            print(f"Failed to load model (local): {e}")
+            print(f"Failed to load model (local): {e}", file=sys.stderr)
             sys.exit(1)
 
     # Otherwise, try from_pretrained (Hub); exit on failure
@@ -95,10 +95,10 @@ def load_model(args) -> VoxCPM:
             lora_config=lora_config,
             lora_weights_path=lora_weights_path,
         )
-        print("Model loaded (from_pretrained).")
+        print("Model loaded (from_pretrained).", file=sys.stderr)
         return model
     except Exception as e:
-        print(f"Failed to load model (from_pretrained): {e}")
+        print(f"Failed to load model (from_pretrained): {e}", file=sys.stderr)
         sys.exit(1)
 
 
@@ -106,15 +106,15 @@ def cmd_clone(args):
     """Voice cloning command."""
     # Validate inputs
     if not args.text:
-        print("Error: Please provide text to synthesize (--text)")
+        print("Error: Please provide text to synthesize (--text)", file=sys.stderr)
         sys.exit(1)
     
     if not args.prompt_audio:
-        print("Error: Voice cloning requires a reference audio (--prompt-audio)")
+        print("Error: Voice cloning requires a reference audio (--prompt-audio)", file=sys.stderr)
         sys.exit(1)
         
     if not args.prompt_text:
-        print("Error: Voice cloning requires a reference text (--prompt-text)")
+        print("Error: Voice cloning requires a reference text (--prompt-text)", file=sys.stderr)
         sys.exit(1)
     
     # Validate files
@@ -125,9 +125,9 @@ def cmd_clone(args):
     model = load_model(args)
     
     # Generate audio
-    print(f"Synthesizing text: {args.text}")
-    print(f"Reference audio: {prompt_audio_path}")
-    print(f"Reference text: {args.prompt_text}")
+    print(f"Synthesizing text: {args.text}", file=sys.stderr)
+    print(f"Reference audio: {prompt_audio_path}", file=sys.stderr)
+    print(f"Reference text: {args.prompt_text}", file=sys.stderr)
     
     audio_array = model.generate(
         text=args.text,
@@ -141,25 +141,25 @@ def cmd_clone(args):
     
     # Save audio
     sf.write(str(output_path), audio_array, model.tts_model.sample_rate)
-    print(f"Saved audio to: {output_path}")
+    print(f"Saved audio to: {output_path}", file=sys.stderr)
     
     # Stats
     duration = len(audio_array) / model.tts_model.sample_rate
-    print(f"Duration: {duration:.2f}s")
+    print(f"Duration: {duration:.2f}s", file=sys.stderr)
 
 
 def cmd_synthesize(args):
     """Direct TTS synthesis command."""
     # Validate inputs
     if not args.text:
-        print("Error: Please provide text to synthesize (--text)")
+        print("Error: Please provide text to synthesize (--text)", file=sys.stderr)
         sys.exit(1)
     # Validate output path
     output_path = validate_output_path(args.output)
     # Load model
     model = load_model(args)
     # Generate audio
-    print(f"Synthesizing text: {args.text}")
+    print(f"Synthesizing text: {args.text}", file=sys.stderr)
     
     audio_array = model.generate(
         text=args.text,
@@ -173,11 +173,11 @@ def cmd_synthesize(args):
     
     # Save audio
     sf.write(str(output_path), audio_array, model.tts_model.sample_rate)
-    print(f"Saved audio to: {output_path}")
+    print(f"Saved audio to: {output_path}", file=sys.stderr)
     
     # Stats
     duration = len(audio_array) / model.tts_model.sample_rate
-    print(f"Duration: {duration:.2f}s")
+    print(f"Duration: {duration:.2f}s", file=sys.stderr)
 
 
 def cmd_batch(args):
@@ -191,12 +191,12 @@ def cmd_batch(args):
         with open(input_file, 'r', encoding='utf-8') as f:
             texts = [line.strip() for line in f if line.strip()]
     except Exception as e:
-        print(f"Failed to read input file: {e}")
+        print(f"Failed to read input file: {e}", file=sys.stderr)
         sys.exit(1)
     if not texts:
-        print("Error: Input file is empty or contains no valid lines")
+        print("Error: Input file is empty or contains no valid lines", file=sys.stderr)
         sys.exit(1)
-    print(f"Found {len(texts)} lines to process")
+    print(f"Found {len(texts)} lines to process", file=sys.stderr)
     
     model = load_model(args)
     prompt_audio_path = None
@@ -205,7 +205,7 @@ def cmd_batch(args):
     
     success_count = 0
     for i, text in enumerate(texts, 1):
-        print(f"\nProcessing {i}/{len(texts)}: {text[:50]}...")
+        print(f"\nProcessing {i}/{len(texts)}: {text[:50]}...", file=sys.stderr)
         
         try:
             audio_array = model.generate(
@@ -221,14 +221,14 @@ def cmd_batch(args):
             sf.write(str(output_file), audio_array, model.tts_model.sample_rate)
             
             duration = len(audio_array) / model.tts_model.sample_rate
-            print(f"  Saved: {output_file} ({duration:.2f}s)")
+            print(f"  Saved: {output_file} ({duration:.2f}s)", file=sys.stderr)
             success_count += 1
             
         except Exception as e:
-            print(f"  Failed: {e}")
+            print(f"  Failed: {e}", file=sys.stderr)
             continue
     
-    print(f"\nBatch finished: {success_count}/{len(texts)} succeeded")
+    print(f"\nBatch finished: {success_count}/{len(texts)} succeeded", file=sys.stderr)
 
 def _build_unified_parser():
     """Build unified argument parser (no subcommands, route by args)."""
@@ -296,14 +296,14 @@ def main():
     # Routing: prefer batch → single (clone/direct)
     if args.input:
         if not args.output_dir:
-            print("Error: Batch mode requires --output-dir")
+            print("Error: Batch mode requires --output-dir", file=sys.stderr)
             parser.print_help()
             sys.exit(1)
         return cmd_batch(args)
 
     # Single-sample mode
     if not args.text or not args.output:
-        print("Error: Single-sample mode requires --text and --output")
+        print("Error: Single-sample mode requires --text and --output", file=sys.stderr)
         parser.print_help()
         sys.exit(1)
 
@@ -316,7 +316,7 @@ def main():
                 args.prompt_text = f.read()
 
         if not args.prompt_audio or not args.prompt_text:
-            print("Error: Voice cloning requires both --prompt-audio and --prompt-text")
+            print("Error: Voice cloning requires both --prompt-audio and --prompt-text", file=sys.stderr)
             sys.exit(1)
         return cmd_clone(args)
 

src/voxcpm/core.py

@@ -1,4 +1,5 @@
 import os
+import sys
 import re
 import tempfile
 import numpy as np
@@ -30,7 +31,7 @@ class VoxCPM:
             lora_weights_path: Path to pre-trained LoRA weights (.pth file or directory
                 containing lora_weights.ckpt). If provided, LoRA weights will be loaded.
         """
-        print(f"voxcpm_model_path: {voxcpm_model_path}, zipenhancer_model_path: {zipenhancer_model_path}, enable_denoiser: {enable_denoiser}")
+        print(f"voxcpm_model_path: {voxcpm_model_path}, zipenhancer_model_path: {zipenhancer_model_path}, enable_denoiser: {enable_denoiser}", file=sys.stderr)
         
         # If lora_weights_path is provided but no lora_config, create a default one
         if lora_weights_path is not None and lora_config is None:
@@ -39,15 +40,15 @@ class VoxCPM:
                 enable_dit=True,
                 enable_proj=False,
             )
-            print(f"Auto-created default LoRAConfig for loading weights from: {lora_weights_path}")
+            print(f"Auto-created default LoRAConfig for loading weights from: {lora_weights_path}", file=sys.stderr)
         
         self.tts_model = VoxCPMModel.from_local(voxcpm_model_path, optimize=optimize, lora_config=lora_config)
         
         # Load LoRA weights if path is provided
         if lora_weights_path is not None:
-            print(f"Loading LoRA weights from: {lora_weights_path}")
+            print(f"Loading LoRA weights from: {lora_weights_path}", file=sys.stderr)
             loaded_keys, skipped_keys = self.tts_model.load_lora_weights(lora_weights_path)
-            print(f"Loaded {len(loaded_keys)} LoRA parameters, skipped {len(skipped_keys)}")
+            print(f"Loaded {len(loaded_keys)} LoRA parameters, skipped {len(skipped_keys)}", file=sys.stderr)
         
         self.text_normalizer = None
         if enable_denoiser and zipenhancer_model_path is not None:
@@ -56,7 +57,7 @@ class VoxCPM:
         else:
             self.denoiser = None
         if optimize:
-            print("Warm up VoxCPMModel...")
+            print("Warm up VoxCPMModel...", file=sys.stderr)
             self.tts_model.generate(
                 target_text="Hello, this is the first test sentence.",
                 max_len=10,

src/voxcpm/model/voxcpm.py

@@ -19,6 +19,7 @@ limitations under the License.
 """
 
 import os
+import sys
 from typing import Tuple, Union, Generator, List, Optional
 
 import torch
@@ -120,7 +121,7 @@ class VoxCPMModel(nn.Module):
                 self.device = "mps"
             else:
                 self.device = "cpu"
-        print(f"Running on device: {self.device}, dtype: {self.config.dtype}")
+        print(f"Running on device: {self.device}, dtype: {self.config.dtype}", file=sys.stderr)
 
         # Text-Semantic LM
         self.base_lm = MiniCPMModel(config.lm_config)
@@ -228,7 +229,7 @@ class VoxCPMModel(nn.Module):
             self.feat_encoder = torch.compile(self.feat_encoder, mode="reduce-overhead", fullgraph=True)
             self.feat_decoder.estimator = torch.compile(self.feat_decoder.estimator, mode="reduce-overhead", fullgraph=True)
         except Exception as e:
-            print(f"Warning: torch.compile disabled - {e}")
+            print(f"Warning: torch.compile disabled - {e}", file=sys.stderr)
         return self
 
     def forward(
@@ -459,7 +460,7 @@ class VoxCPMModel(nn.Module):
                 latent_pred, pred_audio_feat = next(inference_result)
                 if retry_badcase:
                     if pred_audio_feat.shape[0] >= target_text_length * retry_badcase_ratio_threshold:
-                        print(f"  Badcase detected, audio_text_ratio={pred_audio_feat.shape[0] / target_text_length}, retrying...")
+                        print(f"  Badcase detected, audio_text_ratio={pred_audio_feat.shape[0] / target_text_length}, retrying...", file=sys.stderr)
                         retry_badcase_times += 1
                         continue
                     else:
@@ -683,7 +684,7 @@ class VoxCPMModel(nn.Module):
                 latent_pred, pred_audio_feat = next(inference_result)
                 if retry_badcase:
                     if pred_audio_feat.shape[0] >= target_text_length * retry_badcase_ratio_threshold:
-                        print(f"  Badcase detected, audio_text_ratio={pred_audio_feat.shape[0] / target_text_length}, retrying...")
+                        print(f"  Badcase detected, audio_text_ratio={pred_audio_feat.shape[0] / target_text_length}, retrying...", file=sys.stderr)
                         retry_badcase_times += 1
                         continue
                     else:
@@ -868,10 +869,10 @@ class VoxCPMModel(nn.Module):
         pytorch_model_path = os.path.join(path, "pytorch_model.bin")
         
         if os.path.exists(safetensors_path) and SAFETENSORS_AVAILABLE:
-            print(f"Loading model from safetensors: {safetensors_path}")
+            print(f"Loading model from safetensors: {safetensors_path}", file=sys.stderr)
             model_state_dict = load_file(safetensors_path)
         elif os.path.exists(pytorch_model_path):
-            print(f"Loading model from pytorch_model.bin: {pytorch_model_path}")
+            print(f"Loading model from pytorch_model.bin: {pytorch_model_path}", file=sys.stderr)
             checkpoint = torch.load(
                 pytorch_model_path,
                 map_location="cpu",

src/voxcpm/training/tracker.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import contextlib
+import sys
 import time
 from pathlib import Path
 from typing import Dict, Optional
@@ -36,7 +37,7 @@ class TrainingTracker:
     # ------------------------------------------------------------------ #
     def print(self, message: str):
         if self.rank == 0:
-            print(message, flush=True)
+            print(message, flush=True, file=sys.stderr)
             if self.log_file:
                 with self.log_file.open("a", encoding="utf-8") as f:
                     f.write(message + "\n")

test_concurrent.py

@@ -0,0 +1,365 @@
+import asyncio
+import aiohttp
+import time
+import statistics
+import csv
+from typing import List, Dict
+from datetime import datetime
+
+# ---------------- 基本配置 ----------------
+BASE_URL = "http://localhost:8880/generate_tts"
+
+REQUEST_TEMPLATE = {
+    "text": "哇，你这个新买的T-shirt好cool啊！是哪个brand的？周末我们去新开的mall里那家Starbucks喝杯coffee吧？我听说他们的new season限定款Latte很OK。"
+}
+
+# REQUEST_TEMPLATE = {
+#     "text": "澳门在哪里啊",
+#     "cfg_value": "2.0",
+#     "inference_timesteps": "10",
+#     "normalize": "true",
+#     "denoise": "true",
+#     "prompt_text": "澳门有乜嘢好食嘅？"
+# }
+
+# # 音频路径（确保文件存在）
+# PROMPT_WAV_PATH = "/home/verachen/Music/voice/2food.wav"
+
+# headers 一般不必指定 multipart，会自动设置
+DEFAULT_HEADERS = {}
+
+
+# ---------------- 请求逻辑 ----------------
+async def tts_request(session: aiohttp.ClientSession, request_id: int) -> float:
+    """执行一次请求，返回耗时（秒）"""
+    start = time.perf_counter()
+    try:
+        form = aiohttp.FormData()
+        for k, v in REQUEST_TEMPLATE.items():
+            form.add_field(k, v)     
+        # form.add_field(
+        #     "prompt_wav",
+        #     open(PROMPT_WAV_PATH, "rb"),
+        #     filename="2food.wav",
+        #     content_type="audio/wav"
+        # )           
+           
+        async with session.post(BASE_URL, data=form, headers=DEFAULT_HEADERS) as resp:
+            await resp.read()
+            if resp.status != 200:
+                raise RuntimeError(f"HTTP {resp.status}")
+    except Exception as e:
+        print(f"[请求 {request_id}] 出错: {e}")
+        return -1
+    return time.perf_counter() - start
+
+
+# ---------------- 并发测试核心 ----------------
+async def benchmark(concurrency: int, total_requests: int) -> Dict:
+    """在指定并发下发起多次请求，统计性能指标"""
+    timings = []
+    errors = 0
+
+    conn = aiohttp.TCPConnector(limit=0, force_close=False)
+    timeout = aiohttp.ClientTimeout(total=None)
+    async with aiohttp.ClientSession(connector=conn, timeout=timeout) as session:
+        sem = asyncio.Semaphore(concurrency)
+
+        async def worker(i):
+            nonlocal errors
+            async with sem:
+                t = await tts_request(session, i)
+                if t > 0:
+                    timings.append(t)
+                else:
+                    errors += 1
+
+        tasks = [asyncio.create_task(worker(i)) for i in range(total_requests)]
+        await asyncio.gather(*tasks)
+
+    succ = len(timings)
+    total_time = sum(timings)
+    result = {
+        "concurrency": concurrency,
+        "total_requests": total_requests,
+        "successes": succ,
+        "errors": errors,
+    }
+
+    if succ > 0:
+        result.update({
+            "min": min(timings),
+            "max": max(timings),
+            "avg": statistics.mean(timings),
+            "median": statistics.median(timings),
+            "qps": succ / total_time if total_time > 0 else 0
+        })
+    return result
+
+
+# ---------------- CSV 写入 ----------------
+def write_to_csv(filename: str, data: List[Dict]):
+    """把测试结果写入 CSV 文件"""
+    fieldnames = ["concurrency", "total_requests", "successes", "errors",
+                  "min", "max", "avg", "median", "qps"]
+    with open(filename, "w", newline="", encoding="utf-8") as f:
+        writer = csv.DictWriter(f, fieldnames=fieldnames)
+        writer.writeheader()
+        for row in data:
+            writer.writerow(row)
+    print(f"\n✅ 测试结果已保存到: {filename}")
+
+
+# ---------------- 主流程 ----------------
+async def run_tests(concurrency_list: List[int], total_requests: dict):
+    print("开始压测接口:", BASE_URL)
+    results = []
+    for c in concurrency_list:
+        print(f"\n=== 并发数: {c} ===")
+        res = await benchmark(c, total_requests[c])
+        results.append(res)
+
+        print(f"请求总数: {res['total_requests']} | 成功: {res['successes']} | 失败: {res['errors']}")
+        if "avg" in res:
+            print(f"耗时 (s) → min={res['min']:.3f}, avg={res['avg']:.3f}, median={res['median']:.3f}, max={res['max']:.3f}")
+            print(f"近似 QPS: {res['qps']:.2f}")
+        else:
+            print("所有请求均失败")
+
+    # 保存结果
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M")
+    csv_filename = f"tts_benchmark_{timestamp}.csv"
+    write_to_csv(csv_filename, results)
+
+
+if __name__ == "__main__":
+    # 你可以根据机器能力调整这两个参数
+    concurrency_list = [1, 5, 10, 20, 50, 100, 150, 200]  # 并发测试范围
+    total_requests = {
+        1: 50,
+        5: 100,
+        10: 200,
+        20: 400,
+        50: 1000,
+        100: 2000,
+        150: 3000,
+        200: 4000,
+        300: 6000,
+        500: 10000
+    }                                  # 每个并发等级下的请求数
+    asyncio.run(run_tests(concurrency_list, total_requests))
+
+
+# import asyncio
+# import aiohttp
+# import time
+# import statistics
+# from typing import List, Dict
+
+# # 新的接口 URL
+# BASE_URL = "http://127.0.0.1:8880/generate_tts"
+
+# # # 固定参数（不包括音频）
+# # FORM_PARAMS = {
+# #     "text": "澳门在哪里啊",
+# #     "cfg_value": "2.0",
+# #     "inference_timesteps": "10",
+# #     "normalize": "true",
+# #     "denoise": "true",
+# #     "prompt_text": "澳门有乜嘢好食嘅？"
+# # }
+# FORM_PARAMS = {
+#     "text": "澳门在哪里啊"
+# }
+
+
+# # 音频路径（确保文件存在）
+# # PROMPT_WAV_PATH = "/home/verachen/Music/voice/2food.wav"
+
+# # headers 一般不必指定 multipart，会自动设置
+# DEFAULT_HEADERS = {}
+
+
+# async def tts_request(session: aiohttp.ClientSession) -> float:
+#     """
+#     发起一次 multipart/form-data 格式的 TTS 请求，返回耗时。
+#     """
+#     start = time.perf_counter()
+#     form = aiohttp.FormData()
+#     for k, v in FORM_PARAMS.items():
+#         form.add_field(k, v)
+#     # form.add_field(
+#     #     "prompt_wav",
+#     #     open(PROMPT_WAV_PATH, "rb"),
+#     #     filename="2food.wav",
+#     #     content_type="audio/wav"
+#     # )
+
+#     async with session.post(BASE_URL, data=form, headers=DEFAULT_HEADERS) as resp:
+#         data = await resp.read()
+#         if resp.status != 200:
+#             raise RuntimeError(f"HTTP {resp.status}, body: {data[:200]!r}")
+
+#     elapsed = time.perf_counter() - start
+#     return elapsed
+
+
+# async def benchmark(concurrency: int, total_requests: int) -> Dict:
+#     timings: List[float] = []
+#     errors = 0
+
+#     conn = aiohttp.TCPConnector(limit=0)
+#     timeout = aiohttp.ClientTimeout(total=None)
+
+#     async with aiohttp.ClientSession(connector=conn, timeout=timeout) as session:
+#         sem = asyncio.Semaphore(concurrency)
+
+#         async def worker():
+#             nonlocal errors
+#             async with sem:
+#                 try:
+#                     t = await tts_request(session)
+#                     timings.append(t)
+#                 except Exception as e:
+#                     errors += 1
+#                     print("请求出错：", e)
+
+#         tasks = [asyncio.create_task(worker()) for _ in range(total_requests)]
+#         await asyncio.gather(*tasks)
+
+#     succ = len(timings)
+#     total_time = sum(timings) if timings else 0.0
+
+#     result = {
+#         "concurrency": concurrency,
+#         "total_requests": total_requests,
+#         "successes": succ,
+#         "errors": errors,
+#     }
+#     if succ > 0:
+#         result.update({
+#             "min": min(timings),
+#             "max": max(timings),
+#             "avg": statistics.mean(timings),
+#             "median": statistics.median(timings),
+#             "qps": succ / total_time if total_time > 0 else None
+#         })
+#     return result
+
+
+# async def run_tests(concurrency_list: List[int], total_requests: int):
+#     print("开始并发测试，目标接口：", BASE_URL)
+#     for c in concurrency_list:
+#         print(f"\n--- 并发 = {c} ---")
+#         res = await benchmark(c, total_requests)
+#         print("总请求：", res["total_requests"])
+#         print("成功：", res["successes"], "失败：", res["errors"])
+#         if "avg" in res:
+#             print(f"耗时 min / avg / median / max = "
+#                   f"{res['min']:.3f} / {res['avg']:.3f} / {res['median']:.3f} / {res['max']:.3f} 秒")
+#             print(f"近似 QPS = {res['qps']:.2f}")
+#         else:
+#             print("所有请求都失败了")
+
+
+# if __name__ == "__main__":
+#     # 你可以调整这些参数
+#     concurrency_list = [200]     # 要试的并发数列表
+#     total_requests = 100                      # 每个并发等级下总请求数
+
+#     asyncio.run(run_tests(concurrency_list, total_requests))
+
+
+# import multiprocessing as mp
+# import requests
+# import time
+# import statistics
+# from typing import Dict, List
+
+# BASE_URL = "http://127.0.0.1:8880/generate_tts"
+# FORM_PARAMS = {
+#     "text": "澳门在哪里啊",
+#     "cfg_value": "2.0",
+#     "inference_timesteps": "10",
+#     "normalize": "true",
+#     "denoise": "true",
+#     "prompt_text": "澳门有乜嘢好食嘅？"
+# }
+# PROMPT_WAV_PATH = "/home/verachen/Music/voice/2food.wav"
+
+# def single_request() -> float:
+#     """用 requests 同步发一次 multipart/form-data 请求，返回耗时（秒）或抛异常。"""
+#     files = {
+#         "prompt_wav": ("2food.wav", open(PROMPT_WAV_PATH, "rb"), "audio/wav")
+#     }
+#     data = FORM_PARAMS.copy()
+#     start = time.perf_counter()
+#     resp = requests.post(BASE_URL, data=data, files=files, timeout=60)
+#     elapsed = time.perf_counter() - start
+#     if resp.status_code != 200:
+#         raise RuntimeError(f"HTTP {resp.status_code}, body: {resp.text[:200]!r}")
+#     return elapsed
+
+# def worker_task(num_requests: int, return_list: mp.Manager().list, err_list: mp.Manager().list):
+#     """子进程做 num_requests 次请求，将各次耗时记录到 return_list（共享 list），错误次数记录到 err_list。"""
+#     for _ in range(num_requests):
+#         try:
+#             t = single_request()
+#             return_list.append(t)
+#         except Exception as e:
+#             err_list.append(str(e))
+
+# def run_multiproc(concurrency: int, total_requests: int) -> Dict:
+#     """
+#     用多个进程模拟并发：
+#     - 每个进程发 total_requests／concurrency 次请求（向下取整或略分配）
+#     - 或者简单地让每个进程跑 total_requests 次（更激进）
+#     """
+#     manager = mp.Manager()
+#     times = manager.list()
+#     errs = manager.list()
+
+#     procs = []
+#     # 任务分配：每个子进程跑一部分请求
+#     per = total_requests // concurrency
+#     if per < 1:
+#         per = 1
+
+#     for i in range(concurrency):
+#         p = mp.Process(target=worker_task, args=(per, times, errs))
+#         p.start()
+#         procs.append(p)
+
+#     for p in procs:
+#         p.join()
+
+#     timings = list(times)
+#     errors = list(errs)
+#     succ = len(timings)
+#     total_time = sum(timings) if timings else 0.0
+
+#     ret = {
+#         "concurrency": concurrency,
+#         "total_requests": total_requests,
+#         "successes": succ,
+#         "errors": len(errors),
+#     }
+#     if succ > 0:
+#         ret.update({
+#             "min": min(timings),
+#             "max": max(timings),
+#             "avg": statistics.mean(timings),
+#             "median": statistics.median(timings),
+#             "qps": succ / total_time if total_time > 0 else None
+#         })
+#     return ret
+
+# if __name__ == "__main__":
+#     concurrency = 4
+#     total_requests = 40
+#     print("开始多进程并发测试")
+#     result = run_multiproc(concurrency, total_requests)
+#     print(result)
+
+
+

test_streaming.py

@@ -0,0 +1,34 @@
+import asyncio
+import httpx
+import time
+
+async def test_streaming():
+    url = "http://localhost:8880/generate_tts_streaming"
+    data = {
+        "text": "你好，这是一段流式输出的测试音频。",
+        "cfg_value": "2.0",
+        "inference_timesteps": "10",
+        "do_normalize": "True",
+        "denoise": "True"
+    }
+    
+    start_time = time.time()
+    first_byte_received = False
+
+    async with httpx.AsyncClient() as client:
+        # 模拟文件上传（如果有 prompt_wav）
+        # files = {'prompt_wav': open('test.wav', 'rb')}
+        
+        async with client.stream("POST", url, data=data) as response:
+            print(f"状态码: {response.status_code}")
+            
+            async for chunk in response.aiter_bytes():
+                if not first_byte_received:
+                    ttfb = time.time() - start_time
+                    print(f"🚀 首包到达 (TTFB): {ttfb:.4f} 秒")
+                    first_byte_received = True
+                
+                print(f"收到数据块: {len(chunk)} 字节")
+
+if __name__ == "__main__":
+    asyncio.run(test_streaming())