made changes

Dockerfile

@@ -35,9 +35,9 @@ COPY requirements.txt .
 # Install Python dependencies with proper error handling
 RUN pip install --no-cache-dir --upgrade pip setuptools wheel && \
     # Install ONNX Runtime CPU version specifically
-    pip install --no-cache-dir onnxruntime==1.16.3 && \
+    # pip install --no-cache-dir onnxruntime==1.16.3 && \
     # Fix executable stack issue
-    find /usr/local/lib/python*/site-packages/onnxruntime -name "*.so" -exec execstack -c {} \; 2>/dev/null || true && \
+    # find /usr/local/lib/python*/site-packages/onnxruntime -name "*.so" -exec execstack -c {} \; 2>/dev/null || true && \
     # Install other requirements
     pip install --no-cache-dir -r requirements.txt
 
@@ -71,9 +71,9 @@ ENV PYTHONPATH=/app \
     HF_HUB_CACHE=/app/model_cache \
     FONTCONFIG_PATH=/tmp/fontconfig \
     # Critical ONNX Runtime fixes for containers
-    ORT_DYLIB_DEFAULT_OPTIONS=DisableExecutablePageAllocator=1 \
-    ONNXRUNTIME_EXECUTION_PROVIDERS=CPUExecutionProvider \
-    ORT_DISABLE_TLS_ARENA=1 \
+    # ORT_DYLIB_DEFAULT_OPTIONS=DisableExecutablePageAllocator=1 \
+    # ONNXRUNTIME_EXECUTION_PROVIDERS=CPUExecutionProvider \
+    # ORT_DISABLE_TLS_ARENA=1 \
     CTRANSLATE2_FORCE_CPU_ISA=generic \
     # Threading and memory optimizations
     TF_CPP_MIN_LOG_LEVEL=2 \

model_preloader.py

@@ -30,30 +30,40 @@ from rich.text import Text
 import psutil
 
 # CRITICAL: Configure ONNX Runtime BEFORE any ML library imports
-import os
-os.environ.update({
-    'ORT_DYLIB_DEFAULT_OPTIONS': 'DisableExecutablePageAllocator=1',
-    'ONNXRUNTIME_EXECUTION_PROVIDERS': 'CPUExecutionProvider',
-    'ORT_DISABLE_TLS_ARENA': '1',
-    'OMP_NUM_THREADS': '1',
-    'MKL_NUM_THREADS': '1',
-    'NUMBA_NUM_THREADS': '1',
-    'TF_ENABLE_ONEDNN_OPTS': '0',
-    'TOKENIZERS_PARALLELISM': 'false',
-    'MALLOC_ARENA_MAX': '2'
-})
+# import os
+# os.environ.update({
+#     'ORT_DYLIB_DEFAULT_OPTIONS': 'DisableExecutablePageAllocator=1',
+#     'ONNXRUNTIME_EXECUTION_PROVIDERS': 'CPUExecutionProvider',
+#     'ORT_DISABLE_TLS_ARENA': '1',
+#     'OMP_NUM_THREADS': '1',
+#     'MKL_NUM_THREADS': '1',
+#     'NUMBA_NUM_THREADS': '1',
+#     'TF_ENABLE_ONEDNN_OPTS': '0',
+#     'TOKENIZERS_PARALLELISM': 'false',
+#     'MALLOC_ARENA_MAX': '2'
+# })
 
 # Import ONNX Runtime with error suppression
-try:
-    import warnings
-    warnings.filterwarnings("ignore", category=UserWarning, module="onnxruntime")
-    import onnxruntime as ort
-    # Force CPU provider only
-    ort.set_default_logger_severity(3)  # ERROR level only
-except ImportError:
-    pass
-except Exception as e:
-    print(f"ONNX Runtime warning (expected in containers): {e}")
+# try:
+#     import warnings
+#     warnings.filterwarnings("ignore", category=UserWarning, module="onnxruntime")
+#     import onnxruntime as ort
+#     # Force CPU provider only
+#     ort.set_default_logger_severity(3)  # ERROR level only
+# except ImportError:
+#     pass
+# except Exception as e:
+#     print(f"ONNX Runtime warning (expected in containers): {e}")
+
+# CRITICAL: Set environment variables BEFORE importing any ML libraries
+# This fixes the ONNX Runtime executable stack issue in containers
+# os.environ.update({
+#     'ORT_DYLIB_DEFAULT_OPTIONS': 'DisableExecutablePageAllocator=1',
+#     'ONNXRUNTIME_EXECUTION_PROVIDERS': 'CPUExecutionProvider',
+#     'OMP_NUM_THREADS': '1',
+#     'TF_ENABLE_ONEDNN_OPTS': '0',
+#     'TOKENIZERS_PARALLELISM': 'false'
+# })
 
 # Add src directory to path
 sys.path.insert(0, os.path.join(os.path.dirname(__file__), 'src'))
@@ -64,16 +74,6 @@ logger = logging.getLogger(__name__)
 
 console = Console()
 
-# CRITICAL: Set environment variables BEFORE importing any ML libraries
-# This fixes the ONNX Runtime executable stack issue in containers
-os.environ.update({
-    'ORT_DYLIB_DEFAULT_OPTIONS': 'DisableExecutablePageAllocator=1',
-    'ONNXRUNTIME_EXECUTION_PROVIDERS': 'CPUExecutionProvider',
-    'OMP_NUM_THREADS': '1',
-    'TF_ENABLE_ONEDNN_OPTS': '0',
-    'TOKENIZERS_PARALLELISM': 'false'
-})
-
 class ModelPreloader:
     """Comprehensive model preloader with enhanced local cache detection."""
     
@@ -397,20 +397,20 @@ class ModelPreloader:
         except Exception as e:
             logger.warning(f"Error saving cache for {model_key}: {e}")
     
-    def load_pyannote_pipeline(self) -> Optional[Pipeline]:
+    def load_pyannote_pipeline(self, task_id: str) -> Optional[Pipeline]:
         """Load pyannote speaker diarization pipeline with container-safe settings."""
         try:
             console.print(f"[yellow]Loading pyannote.audio pipeline...[/yellow]")
             
             # Fix ONNX Runtime libraries first
-            try:
-                import subprocess
-                subprocess.run([
-                    'find', '/usr/local/lib/python*/site-packages/onnxruntime', 
-                    '-name', '*.so', '-exec', 'execstack', '-c', '{}', ';'
-                ], capture_output=True, timeout=10, stderr=subprocess.DEVNULL)
-            except:
-                pass
+            # try:
+            #     import subprocess
+            #     subprocess.run([
+            #         'find', '/usr/local/lib/python*/site-packages/onnxruntime', 
+            #         '-name', '*.so', '-exec', 'execstack', '-c', '{}', ';'
+            #     ], capture_output=True, timeout=10, stderr=subprocess.DEVNULL)
+            # except:
+            #     pass
             
             # Check for HuggingFace token
             hf_token = os.getenv('HUGGINGFACE_TOKEN') or os.getenv('HF_TOKEN')
@@ -429,7 +429,7 @@ class ModelPreloader:
             os.environ['ORT_LOGGING_LEVEL'] = '3'  # ERROR only
             
             # Disable other verbose logging
-            logging.getLogger('onnxruntime').setLevel(logging.ERROR)
+            # logging.getLogger('onnxruntime').setLevel(logging.ERROR)
             logging.getLogger('transformers').setLevel(logging.ERROR)
             
             try:
@@ -453,28 +453,28 @@ class ModelPreloader:
                 warnings.filters[:] = old_warning_filters
             
         except Exception as e:
-            error_msg = str(e).lower()
-            if "executable stack" in error_msg or "onnxruntime" in error_msg:
-                console.print("[yellow]ONNX Runtime container warning (attempting workaround)...[/yellow]")
+            # error_msg = str(e).lower()
+            # if "executable stack" in error_msg or "onnxruntime" in error_msg:
+            #     console.print("[yellow]ONNX Runtime container warning (attempting workaround)...[/yellow]")
                 
-                # Try alternative approach - load without ONNX-dependent components
-                try:
-                    # Try loading with CPU-only execution providers
-                    import onnxruntime as ort
-                    ort.set_default_logger_severity(4)  # FATAL only
+            #     # Try alternative approach - load without ONNX-dependent components
+            #     try:
+            #         # Try loading with CPU-only execution providers
+            #         import onnxruntime as ort
+            #         ort.set_default_logger_severity(4)  # FATAL only
                     
-                    pipeline = Pipeline.from_pretrained(
-                        "pyannote/speaker-diarization-3.1",
-                        use_auth_token=hf_token,
-                        cache_dir=str(self.cache_dir / "pyannote")
-                    )
-                    console.print(f"[green]SUCCESS: pyannote.audio loaded with workaround[/green]")
-                    return pipeline
+            #         pipeline = Pipeline.from_pretrained(
+            #             "pyannote/speaker-diarization-3.1",
+            #             use_auth_token=hf_token,
+            #             cache_dir=str(self.cache_dir / "pyannote")
+            #         )
+            #         console.print(f"[green]SUCCESS: pyannote.audio loaded with workaround[/green]")
+            #         return pipeline
                     
-                except Exception as e2:
-                    console.print(f"[red]ERROR: All pyannote loading methods failed: {e2}[/red]")
-            else:
-                console.print(f"[red]ERROR: Failed to load pyannote.audio pipeline: {e}[/red]")
+            #     except Exception as e2:
+            #         console.print(f"[red]ERROR: All pyannote loading methods failed: {e2}[/red]")
+            # else:
+            #     console.print(f"[red]ERROR: Failed to load pyannote.audio pipeline: {e}[/red]")
             
             logger.error(f"Pyannote loading failed: {e}")
             return None

requirements.txt

@@ -4,7 +4,7 @@ torchaudio==2.0.2
 torchvision==0.15.2
 
 # Keep regular ONNX Runtime with container-safe environment variables
-onnxruntime==1.15.1
+# onnxruntime==1.15.1
 
 # Audio processing
 pyannote.audio==3.1.1

startup.py

@@ -5,21 +5,21 @@ Handles model preloading and graceful fallbacks for containerized environments.
 """
 
 # Suppress ONNX Runtime warnings BEFORE any imports
-import warnings
-warnings.filterwarnings("ignore", message=".*executable stack.*")
-warnings.filterwarnings("ignore", category=UserWarning, module="onnxruntime")
+# import warnings
+# warnings.filterwarnings("ignore", message=".*executable stack.*")
+# warnings.filterwarnings("ignore", category=UserWarning, module="onnxruntime")
 
 import os
 import subprocess
 import sys
 import logging
 
-# Set critical environment variables immediately
-os.environ.update({
-    'ORT_DYLIB_DEFAULT_OPTIONS': 'DisableExecutablePageAllocator=1',
-    'ONNXRUNTIME_EXECUTION_PROVIDERS': 'CPUExecutionProvider',
-    'ORT_DISABLE_TLS_ARENA': '1'
-})
+# # Set critical environment variables immediately
+# os.environ.update({
+#     'ORT_DYLIB_DEFAULT_OPTIONS': 'DisableExecutablePageAllocator=1',
+#     'ONNXRUNTIME_EXECUTION_PROVIDERS': 'CPUExecutionProvider',
+#     'ORT_DISABLE_TLS_ARENA': '1'
+# })
 
 # Configure logging
 logging.basicConfig(
@@ -71,39 +71,39 @@ def preload_models():
         logger.info('✅ Model preloader module found')
         
         # Set comprehensive environment variables for ONNX Runtime
-        env = os.environ.copy()
-        env.update({
-            'ORT_DYLIB_DEFAULT_OPTIONS': 'DisableExecutablePageAllocator=1',
-            'ONNXRUNTIME_EXECUTION_PROVIDERS': 'CPUExecutionProvider',
-            'ORT_DISABLE_TLS_ARENA': '1',
-            'TF_ENABLE_ONEDNN_OPTS': '0',
-            'OMP_NUM_THREADS': '1',
-            'MKL_NUM_THREADS': '1',
-            'NUMBA_NUM_THREADS': '1',
-            'TOKENIZERS_PARALLELISM': 'false',
-            'MALLOC_ARENA_MAX': '2',
-            # Additional ONNX Runtime fixes
-            'ONNXRUNTIME_LOG_SEVERITY_LEVEL': '3',
-            'ORT_LOGGING_LEVEL': 'WARNING'
-        })
+        # env = os.environ.copy()
+        # env.update({
+        #     'ORT_DYLIB_DEFAULT_OPTIONS': 'DisableExecutablePageAllocator=1',
+        #     'ONNXRUNTIME_EXECUTION_PROVIDERS': 'CPUExecutionProvider',
+        #     'ORT_DISABLE_TLS_ARENA': '1',
+        #     'TF_ENABLE_ONEDNN_OPTS': '0',
+        #     'OMP_NUM_THREADS': '1',
+        #     'MKL_NUM_THREADS': '1',
+        #     'NUMBA_NUM_THREADS': '1',
+        #     'TOKENIZERS_PARALLELISM': 'false',
+        #     'MALLOC_ARENA_MAX': '2',
+        #     # Additional ONNX Runtime fixes
+        #     'ONNXRUNTIME_LOG_SEVERITY_LEVEL': '3',
+        #     'ORT_LOGGING_LEVEL': 'WARNING'
+        # })
         
-        # Try to fix ONNX Runtime libraries before running preloader
-        try:
-            import subprocess
-            subprocess.run([
-                'find', '/usr/local/lib/python*/site-packages/onnxruntime', 
-                '-name', '*.so', '-exec', 'execstack', '-c', '{}', ';'
-            ], capture_output=True, timeout=30)
-        except:
-            pass  # Continue if execstack fix fails
+        # # Try to fix ONNX Runtime libraries before running preloader
+        # try:
+        #     import subprocess
+        #     subprocess.run([
+        #         'find', '/usr/local/lib/python*/site-packages/onnxruntime', 
+        #         '-name', '*.so', '-exec', 'execstack', '-c', '{}', ';'
+        #     ], capture_output=True, timeout=30)
+        # except:
+        #     pass  # Continue if execstack fix fails
         
         # Try to run the preloader
         result = subprocess.run(
             ['python', 'model_preloader.py'], 
             capture_output=True, 
             text=True, 
-            timeout=300,  # 5 minute timeout
-            env=env
+            timeout=300  # 5 minute timeout
+            # env=env
         )
         
         if result.returncode == 0:
@@ -113,15 +113,15 @@ def preload_models():
             return True
         else:
             logger.warning(f'⚠️ Model preloading failed with return code {result.returncode}')
-            if result.stderr:
-                # Filter out expected ONNX warnings
-                stderr_lines = result.stderr.split('\n')
-                important_errors = [line for line in stderr_lines 
-                                  if 'executable stack' not in line.lower() 
-                                  and 'onnxruntime' not in line.lower() 
-                                  and line.strip()]
-                if important_errors:
-                    logger.warning(f'Important errors: {important_errors[:3]}')
+            # if result.stderr:
+            #     # Filter out expected ONNX warnings
+            #     stderr_lines = result.stderr.split('\n')
+            #     important_errors = [line for line in stderr_lines 
+            #                       if 'executable stack' not in line.lower() 
+            #                       and 'onnxruntime' not in line.lower() 
+            #                       and line.strip()]
+            #     if important_errors:
+            #         logger.warning(f'Important errors: {important_errors[:3]}')
             return False
             
     except subprocess.TimeoutExpired: