Upload Dockerfile with huggingface_hub

Dockerfile

@@ -0,0 +1,76 @@
+FROM pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime
+
+# Set environment variables
+ENV DEBIAN_FRONTEND=noninteractive
+ENV PYTHONUNBUFFERED=1
+ENV HF_HOME=/app/.cache/huggingface
+ENV TRANSFORMERS_CACHE=/app/.cache/huggingface/transformers
+ENV PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:128
+
+# Create necessary directories with proper permissions
+RUN mkdir -p /app/.cache/huggingface/transformers && \
+    chmod -R 777 /app
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential \
+    git \
+    curl \
+    ca-certificates \
+    python3-pip \
+    python3-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+# Create a working directory
+WORKDIR /app
+
+# Install core requirements
+COPY requirements.txt .
+RUN pip3 install --no-cache-dir --upgrade pip && \
+    pip3 install --no-cache-dir -r requirements.txt
+
+# Install basic dependencies specifically for InternViT
+RUN pip3 install --no-cache-dir \
+    transformers==4.37.2 \
+    timm==0.9.11 \
+    accelerate==0.30.0 \
+    safetensors==0.4.1 \
+    einops
+
+# Create a modified test script that can work without flash-attn
+RUN echo 'import torch\nimport os\nimport sys\nimport traceback\nimport gradio as gr\nfrom PIL import Image\nfrom transformers import AutoModel, CLIPImageProcessor\n\nprint("=" * 50)\nprint("INTERNVIT-6B MODEL LOADING TEST (NO FLASH-ATTN)")\nprint("=" * 50)\n\n# System information\nprint(f"Python version: {sys.version}")\nprint(f"PyTorch version: {torch.__version__}")\nprint(f"CUDA available: {torch.cuda.is_available()}")\n\nif torch.cuda.is_available():\n    print(f"CUDA version: {torch.version.cuda}")\n    print(f"GPU count: {torch.cuda.device_count()}")\n    for i in range(torch.cuda.device_count()):\n        print(f"GPU {i}: {torch.cuda.get_device_name(i)}")\n        \n    # Memory info\n    print(f"Total GPU memory: {torch.cuda.get_device_properties(0).total_memory / 1e9:.2f} GB")\n    print(f"Allocated GPU memory: {torch.cuda.memory_allocated() / 1e9:.2f} GB")\n    print(f"Reserved GPU memory: {torch.cuda.memory_reserved() / 1e9:.2f} GB")\nelse:\n    print("CUDA is not available. This is a critical issue for model loading.")\n\n# Create a function to load and test the model\ndef load_and_test_model():\n    try:\n        # Monkey patch to disable flash attention\n        import sys\n        import types\n        \n        # Create a fake flash_attn module\n        flash_attn_module = types.ModuleType("flash_attn")\n        flash_attn_module.__version__ = "0.0.0-disabled"\n        sys.modules["flash_attn"] = flash_attn_module\n        \n        print("\\nNOTE: Created dummy flash_attn module to avoid dependency error")\n        print("This is just for testing basic model loading - some functionality may be disabled")\n        \n        print("\\nLoading model with bfloat16 precision and low_cpu_mem_usage=True...")\n        model = AutoModel.from_pretrained(\n            "OpenGVLab/InternViT-6B-224px",\n            torch_dtype=torch.bfloat16,\n            low_cpu_mem_usage=True,\n            trust_remote_code=True)\n            \n        if torch.cuda.is_available():\n            print("Moving model to CUDA...")\n            model = model.cuda()\n            \n        model.eval()\n        print("✓ Model loaded successfully!")\n        \n        # Now try to process a test image\n        print("\\nLoading image processor...")\n        image_processor = CLIPImageProcessor.from_pretrained("OpenGVLab/InternViT-6B-224px")\n        print("✓ Image processor loaded successfully!")\n        \n        # Create a simple test image\n        print("\\nCreating test image...")\n        test_image = Image.new("RGB", (224, 224), color="red")\n        \n        # Process the test image\n        print("Processing test image...")\n        pixel_values = image_processor(images=test_image, return_tensors="pt").pixel_values\n        if torch.cuda.is_available():\n            pixel_values = pixel_values.to(torch.bfloat16).cuda()\n            \n        # Get model parameters\n        params = sum(p.numel() for p in model.parameters())\n        print(f"Model parameters: {params:,}")\n        \n        # Forward pass\n        print("Running forward pass...")\n        with torch.no_grad():\n            outputs = model(pixel_values)\n            \n        print("✓ Forward pass successful!")\n        print(f"Output shape: {outputs.last_hidden_state.shape}")\n        \n        return f"SUCCESS: Model loaded and test passed!\\nParameters: {params:,}\\nOutput shape: {outputs.last_hidden_state.shape}"\n        \n    except Exception as e:\n        print(f"\\n❌ ERROR: {str(e)}")\n        traceback.print_exc()\n        return f"FAILED: Error loading model or processing image\\nError: {str(e)}"\n\n# Create a simple Gradio interface\ndef create_interface():\n    with gr.Blocks(title="InternViT-6B Test") as demo:\n        gr.Markdown("# InternViT-6B Model Loading Test (without Flash Attention)")\n        gr.Markdown("### This version uses a dummy flash-attn implementation to avoid compilation issues")\n        \n        with gr.Row():\n            test_btn = gr.Button("Test Model Loading")\n            output = gr.Textbox(label="Test Results", lines=10)\n        \n        test_btn.click(fn=load_and_test_model, inputs=[], outputs=output)\n        \n    return demo\n\n# Main function\nif __name__ == "__main__":\n    # Print environment variables\n    print("\\nEnvironment variables:")\n    relevant_vars = ["CUDA_VISIBLE_DEVICES", "NVIDIA_VISIBLE_DEVICES", \n                     "TRANSFORMERS_CACHE", "HF_HOME", "PYTORCH_CUDA_ALLOC_CONF"]\n    for var in relevant_vars:\n        print(f"{var}: {os.environ.get(var, "Not set")}")\n    \n    # Set environment variable for better GPU memory management\n    os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:128"\n    \n    # Create and launch the interface\n    demo = create_interface()\n    demo.launch(share=False, server_name="0.0.0.0")' > /app/no_flash_attn_test.py
+
+# Add a simple script to check GPU status
+RUN echo '#!/bin/bash \n\
+echo "Starting diagnostics..." \n\
+echo "===== System Information =====" \n\
+python3 -c "import sys; print(f\"Python version: {sys.version}\")" \n\
+python3 -c "import torch; print(f\"PyTorch version: {torch.__version__}\")" \n\
+echo "\n===== CUDA Information =====" \n\
+python3 -c "import torch; print(f\"CUDA available: {torch.cuda.is_available()}\")" \n\
+if [ $(python3 -c "import torch; print(torch.cuda.is_available())") = "True" ]; then \n\
+    python3 -c "import torch; print(f\"CUDA version: {torch.version.cuda}\")" \n\
+    python3 -c "import torch; print(f\"GPU count: {torch.cuda.device_count()}\")" \n\
+    python3 -c "import torch; for i in range(torch.cuda.device_count()): print(f\"GPU {i}: {torch.cuda.get_device_name(i)}\")" \n\
+    python3 -c "import torch; print(f\"Total GPU memory: {torch.cuda.get_device_properties(0).total_memory / 1024 / 1024 / 1024:.2f} GB\")" \n\
+fi \n\
+echo "\n===== Package Information =====" \n\
+pip3 list | grep -E "transformers|einops|torch|timm|accelerate|safetensors" \n\
+echo "\n===== NVIDIA System Information =====" \n\
+if command -v nvidia-smi &> /dev/null; then \n\
+    nvidia-smi \n\
+else \n\
+    echo "nvidia-smi not found" \n\
+fi \n\
+echo "\n===== Starting Application =====" \n\
+exec "$@"' > /entrypoint.sh && \
+chmod +x /entrypoint.sh
+
+# Expose port 7860 for Gradio
+EXPOSE 7860
+
+# Use our entrypoint script
+ENTRYPOINT ["/entrypoint.sh"]
+
+# Start the modified application that doesn't require flash-attn
+CMD ["python3", "no_flash_attn_test.py"]