Extract model download logic to utils (#146)

* Extract ensure_model to utils

* migrate existing usage_model refs to utils

* Fix ruff errors

Author: Nash0x7E2

agents-core/vision_agents/core/utils/utils.py

@@ -1,3 +1,4 @@
+import asyncio
 import io
 import logging
 import re
@@ -6,6 +7,7 @@
 from dataclasses import dataclass
 from typing import Dict, Optional
 from PIL import Image
+import httpx
 
 
 # Type alias for markdown file contents: maps filename to file content
@@ -108,3 +110,48 @@ def get_vision_agents_version() -> str:
         return importlib.metadata.version("vision-agents")
     except importlib.metadata.PackageNotFoundError:
         return "unknown"
+
+
+async def ensure_model(path: str, url: str) -> str:
+    """
+    Download a model file asynchronously if it doesn't exist.
+    
+    Args:
+        path: Local path where the model should be saved
+        url: URL to download the model from
+        
+    Returns:
+        The path to the model file
+    """
+
+    logger = logging.getLogger(__name__)
+    if not os.path.exists(path):
+        model_name = os.path.basename(path)
+        logger.info(f"Downloading {model_name}...")
+        
+        try:
+            async with httpx.AsyncClient(timeout=300.0, follow_redirects=True) as client:
+                async with client.stream("GET", url) as response:
+                    response.raise_for_status()
+                    
+                    # Write file in chunks to avoid loading entire file in memory
+                    chunks = []
+                    async for chunk in response.aiter_bytes(chunk_size=8192):
+                        chunks.append(chunk)
+                    
+                    # Write all chunks to file in thread to avoid blocking event loop
+                    def write_file():
+                        with open(path, "wb") as f:
+                            for chunk in chunks:
+                                f.write(chunk)
+                    
+                    await asyncio.to_thread(write_file)
+            
+            logger.info(f"{model_name} downloaded.")
+        except httpx.HTTPError as e:
+            # Clean up partial download on error
+            if os.path.exists(path):
+                os.remove(path)
+            raise RuntimeError(f"Failed to download {model_name}: {e}")
+    
+    return path

plugins/smart_turn/tests/test_smart_turn.py

@@ -9,10 +9,11 @@
 from vision_agents.plugins.smart_turn.smart_turn_detection import (
     SileroVAD,
     SmartTurnDetection,
-    ensure_model,
     SILERO_ONNX_URL,
     SILERO_ONNX_FILENAME,
 )
+from vision_agents.core.utils.utils import ensure_model
+
 import logging
 
 logger = logging.getLogger(__name__)

plugins/smart_turn/vision_agents/plugins/smart_turn/smart_turn_detection.py

@@ -4,7 +4,6 @@
 from dataclasses import dataclass
 from typing import Optional, Any
 
-import httpx
 from getstream.video.rtc.track_util import PcmData, AudioFormat
 import numpy as np
 import onnxruntime as ort
@@ -19,6 +18,7 @@
     TurnStartedEvent,
     TurnEndedEvent,
 )
+from vision_agents.core.utils.utils import ensure_model
 
 import logging
 
@@ -462,47 +462,3 @@ def _predict_speech(self, chunk_f32: np.ndarray) -> float:
 
         # out shape is (1, 1) -> return scalar
         return float(out[0][0])
-
-
-async def ensure_model(path: str, url: str) -> str:
-    """
-    Download a model file asynchronously if it doesn't exist.
-
-    Args:
-        path: Local path where the model should be saved
-        url: URL to download the model from
-
-    Returns:
-        The path to the model file
-    """
-    if not os.path.exists(path):
-        model_name = os.path.basename(path)
-
-        try:
-            async with httpx.AsyncClient(
-                timeout=300.0, follow_redirects=True
-            ) as client:
-                async with client.stream("GET", url) as response:
-                    response.raise_for_status()
-
-                    # Write file in chunks to avoid loading entire file in memory
-                    # Use asyncio.to_thread for blocking file I/O operations
-                    chunks = []
-                    async for chunk in response.aiter_bytes(chunk_size=8192):
-                        chunks.append(chunk)
-
-                    # Write all chunks to file in thread to avoid blocking event loop
-                    def write_file():
-                        with open(path, "wb") as f:
-                            for chunk in chunks:
-                                f.write(chunk)
-
-                    await asyncio.to_thread(write_file)
-
-        except httpx.HTTPError as e:
-            # Clean up partial download on error
-            if os.path.exists(path):
-                os.remove(path)
-            raise RuntimeError(f"Failed to download {model_name}: {e}")
-
-    return path

plugins/vogent/vision_agents/plugins/vogent/vogent_turn_detection.py

@@ -4,7 +4,6 @@
 from dataclasses import dataclass
 from typing import Optional, Any
 
-import httpx
 import numpy as np
 from faster_whisper import WhisperModel
 from getstream.video.rtc.track_util import PcmData, AudioFormat
@@ -17,6 +16,7 @@
     TurnStartedEvent,
     TurnEndedEvent,
 )
+from vision_agents.core.utils.utils import ensure_model
 
 import logging
 
@@ -509,46 +509,3 @@ def predict_speech(self, chunk_f32: np.ndarray) -> float:
         return float(out[0][0])
 
 
-async def ensure_model(path: str, url: str) -> str:
-    """
-    Download a model file asynchronously if it doesn't exist.
-    
-    Args:
-        path: Local path where the model should be saved
-        url: URL to download the model from
-        
-    Returns:
-        The path to the model file
-    """
-    if not os.path.exists(path):
-        model_name = os.path.basename(path)
-        logger.info(f"Downloading {model_name}...")
-        
-        try:
-            async with httpx.AsyncClient(timeout=300.0, follow_redirects=True) as client:
-                async with client.stream("GET", url) as response:
-                    response.raise_for_status()
-                    
-                    # Write file in chunks to avoid loading entire file in memory
-                    chunks = []
-                    async for chunk in response.aiter_bytes(chunk_size=8192):
-                        chunks.append(chunk)
-                    
-                    # Write all chunks to file in thread to avoid blocking event loop
-                    def write_file():
-                        with open(path, "wb") as f:
-                            for chunk in chunks:
-                                f.write(chunk)
-                    
-                    await asyncio.to_thread(write_file)
-            
-            logger.info(f"{model_name} downloaded.")
-        except httpx.HTTPError as e:
-            # Clean up partial download on error
-            if os.path.exists(path):
-                os.remove(path)
-            raise RuntimeError(f"Failed to download {model_name}: {e}")
-    
-    return path
-
-