Committing too many files but the app runs possibly with a new model.

2026-01-16 23:07:45 -05:00
parent 12fd2ef45e
commit cb4488ee58
14 changed files with 61686 additions and 6 deletions
@@ -21,6 +21,14 @@ public class EmbeddingConfiguration
 	public int Dimension { get; set; } = 384;
 	public string ApiToken { get; set; } = string.Empty;
 	public bool UseLocalInference { get; set; } = true;
+	
+	// Predefined model configurations
+	public static class Models
+	{
+		public const string DefaultMiniLM = "sentence-transformers/all-MiniLM-L6-v2";
+		public const string AddressTunedMiniLM = "jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3";
+		public const string AddressTunedMiniLMAlias = "custom-all-MiniLM-L6-v2-address";
+	}
 }

 public class AppSettings
@@ -0,0 +1,112 @@
+# Custom Model Conversion Guide
+
+This document describes how to use a custom embedding model for address embeddings.
+
+## Model Source
+
+The custom model is available at:
+- **HuggingFace**: [jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3](https://huggingface.co/jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3)
+
+This model is a fine-tuned version of all-MiniLM-L6-v2 specifically trained on address data.
+
+## Converting to ONNX Format
+
+Since the model doesn't come with a pre-converted ONNX format, you need to convert it using Python.
+
+### Prerequisites
+
+Install the required packages:
+
+```bash
+pip install optimum[exporters] transformers torch
+```
+
+### Conversion Steps
+
+1. **Run the conversion script**:
+
+   ```bash
+   cd VectorSearchApp/Models
+   python download-convert-model.py
+   ```
+
+   This will:
+   - Download the model from HuggingFace
+   - Convert it to ONNX format using Optimum
+   - Save the model to `Models/custom-model/`
+   - Copy the main model file to `Models/address-embedding-model.onnx`
+
+2. **Update configuration**:
+
+   Edit `VectorSearchApp/appsettings.json`:
+
+   ```json
+   {
+     "Embedding": {
+       "ModelName": "jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3",
+       "Dimension": 384,
+       "ApiToken": "",
+       "UseLocalInference": true
+     }
+   }
+   ```
+
+   Or use the shorter alias:
+
+   ```json
+   {
+     "Embedding": {
+       "ModelName": "custom-all-MiniLM-L6-v2-address",
+       "Dimension": 384,
+       "ApiToken": "",
+       "UseLocalInference": true
+     }
+   }
+   ```
+
+3. **Run the application**:
+
+   ```bash
+   cd VectorSearchApp
+   dotnet run
+   ```
+
+## Output Files
+
+After conversion, the following files will be created:
+
+```
+VectorSearchApp/Models/
+├── custom-model/
+│   ├── config.json
+│   ├── model.onnx
+│   ├── special_tokens_map.json
+│   ├── tokenizer.json
+│   ├── tokenizer_config.json
+│   └── vocab.txt
+└── address-embedding-model.onnx  (copy of model.onnx for easy access)
+```
+
+## Troubleshooting
+
+### CUDA/GPU Support
+
+If you want to use GPU acceleration during conversion:
+
+```python
+from optimum.onnxruntime import ORTModelForFeatureExtraction
+
+model = ORTModelForFeatureExtraction.from_pretrained(
+    model_id,
+    export=True,
+    provider="CUDAExecutionProvider",  # Use CUDA instead of CPU
+)
+```
+
+### Large Model Download
+
+The first conversion may take several minutes as it downloads the full model (~90MB) and tokenizer files.
+
+### Memory Requirements
+
+Conversion requires approximately 4GB of RAM. If you encounter memory issues, try closing other applications.
@@ -0,0 +1,25 @@
+{
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "dtype": "float32",
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 384,
+  "initializer_range": 0.02,
+  "intermediate_size": 1536,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 6,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.57.6",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}
@@ -0,0 +1,37 @@
+{
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "[SEP]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}
@@ -0,0 +1,65 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": true,
+  "extra_special_tokens": {},
+  "mask_token": "[MASK]",
+  "max_length": 128,
+  "model_max_length": 256,
+  "never_split": null,
+  "pad_to_multiple_of": null,
+  "pad_token": "[PAD]",
+  "pad_token_type_id": 0,
+  "padding_side": "right",
+  "sep_token": "[SEP]",
+  "stride": 0,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
+  "unk_token": "[UNK]"
+}
@@ -0,0 +1,35 @@
+#!/usr/bin/env pwsh
+# Download script for custom address embedding model (ONNX format)
+# This script downloads a pre-converted ONNX model if available
+
+$ModelRepo = "jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3"
+$OutputPath = "Models/address-embedding-model.onnx"
+
+Write-Host "Attempting to download pre-converted ONNX model..." -ForegroundColor Cyan
+
+# Try to download from HuggingFace Hub (ONNX format if available)
+$OnnxUrl = "https://huggingface.co/$ModelRepo/resolve/main/onnx/model.onnx"
+
+try {
+    $ProgressPreference = 'SilentlyContinue'
+    
+    if (Get-Command curl -ErrorAction SilentlyContinue) {
+        curl -L -o $OutputPath $OnnxUrl --fail
+    } else {
+        Invoke-WebRequest -Uri $OnnxUrl -OutFile $OutputPath -UseBasicParsing
+    }
+    
+    if (Test-Path $OutputPath) {
+        $size = (Get-Item $OutputPath).Length / 1MB
+        Write-Host "Successfully downloaded model to $OutputPath" -ForegroundColor Green
+        Write-Host "File size: $([math]::Round($size, 2)) MB" -ForegroundColor Gray
+    }
+}
+catch {
+    Write-Host "Could not download pre-converted ONNX model." -ForegroundColor Yellow
+    Write-Host ""
+    Write-Host "The model does not have a pre-converted ONNX format." -ForegroundColor White
+    Write-Host "Please run the Python conversion script instead:" -ForegroundColor White
+    Write-Host "  python download-convert-model.py" -ForegroundColor Gray
+    exit 1
+}
@@ -0,0 +1,177 @@
+#!/usr/bin/env python3
+"""
+Download and convert the custom all-MiniLM-L6-v2 model to ONNX format.
+
+This script:
+1. Downloads the model from HuggingFace (jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3)
+2. Converts it to ONNX format using optimum library
+3. Saves the ONNX model to the Models directory
+
+Requirements:
+    pip install optimum[exporters] transformers torch
+
+Usage:
+    python download-convert-model.py
+"""
+
+import os
+import subprocess
+import sys
+
+def check_requirements():
+    """Check if required packages are installed."""
+    required_packages = [
+        ("optimum", "optimum[exporters]"),
+        ("transformers", "transformers"),
+        ("torch", "torch"),
+    ]
+    
+    missing = []
+    for package_name, install_command in required_packages:
+        try:
+            __import__(package_name.replace("-", "_"))
+            print(f"[OK] {package_name} is installed")
+        except ImportError:
+            missing.append((package_name, install_command))
+    
+    if missing:
+        print("\nMissing required packages. Installing...")
+        for package_name, install_command in missing:
+            print(f"Installing {package_name}...")
+            subprocess.check_call([sys.executable, "-m", "pip", "install", install_command])
+    
+    return True
+
+def download_and_convert_model():
+    """Download model from HuggingFace and convert to ONNX."""
+    from optimum.onnxruntime import ORTModelForFeatureExtraction
+    from transformers import AutoTokenizer
+    
+    model_id = "jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3"
+    output_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), "Models", "custom-model")
+    
+    print(f"\n{'='*60}")
+    print(f"Downloading model: {model_id}")
+    print(f"Output directory: {output_dir}")
+    print(f"{'='*60}\n")
+    
+    # Create output directory
+    os.makedirs(output_dir, exist_ok=True)
+    
+    try:
+        print("Downloading model and tokenizer from HuggingFace...")
+        print("This may take a few minutes on first run...\n")
+        
+        # Download model and tokenizer, then export to ONNX
+        model = ORTModelForFeatureExtraction.from_pretrained(
+            model_id,
+            export=True,
+            provider="CPUExecutionProvider",
+        )
+        
+        tokenizer = AutoTokenizer.from_pretrained(model_id)
+        
+        # Save ONNX model
+        print(f"Saving ONNX model to {output_dir}...")
+        model.save_pretrained(output_dir)
+        tokenizer.save_pretrained(output_dir)
+        
+        print("\n[OK] Model successfully converted to ONNX format!")
+        print(f"\nOutput files:")
+        for f in os.listdir(output_dir):
+            filepath = os.path.join(output_dir, f)
+            size_mb = os.path.getsize(filepath) / (1024 * 1024)
+            print(f"  - {f} ({size_mb:.2f} MB)")
+
+        # Copy the main model file to a simpler location for easy access
+        main_model_file = os.path.join(output_dir, "model.onnx")
+        if os.path.exists(main_model_file):
+            simple_output = os.path.join(os.path.dirname(os.path.abspath(__file__)), "Models", "address-embedding-model.onnx")
+            import shutil
+            shutil.copy(main_model_file, simple_output)
+            print(f"\n[OK] Copied model to: {simple_output}")
+        
+        return True
+        
+    except Exception as e:
+        print(f"\n[ERROR] {e}")
+        import traceback
+        traceback.print_exc()
+        return False
+
+def create_powershell_download_script():
+    """Create a PowerShell script for Windows users who can't run Python."""
+    ps_script = '''#!/usr/bin/env pwsh
+# Download script for custom address embedding model (ONNX format)
+# This script downloads a pre-converted ONNX model if available
+
+$ModelRepo = "jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3"
+$OutputPath = "Models/address-embedding-model.onnx"
+
+Write-Host "Attempting to download pre-converted ONNX model..." -ForegroundColor Cyan
+
+# Try to download from HuggingFace Hub (ONNX format if available)
+$OnnxUrl = "https://huggingface.co/$ModelRepo/resolve/main/onnx/model.onnx"
+
+try {
+    $ProgressPreference = 'SilentlyContinue'
+    
+    if (Get-Command curl -ErrorAction SilentlyContinue) {
+        curl -L -o $OutputPath $OnnxUrl --fail
+    } else {
+        Invoke-WebRequest -Uri $OnnxUrl -OutFile $OutputPath -UseBasicParsing
+    }
+    
+    if (Test-Path $OutputPath) {
+        $size = (Get-Item $OutputPath).Length / 1MB
+        Write-Host "Successfully downloaded model to $OutputPath" -ForegroundColor Green
+        Write-Host "File size: $([math]::Round($size, 2)) MB" -ForegroundColor Gray
+    }
+}
+catch {
+    Write-Host "Could not download pre-converted ONNX model." -ForegroundColor Yellow
+    Write-Host ""
+    Write-Host "The model does not have a pre-converted ONNX format." -ForegroundColor White
+    Write-Host "Please run the Python conversion script instead:" -ForegroundColor White
+    Write-Host "  python download-convert-model.py" -ForegroundColor Gray
+    exit 1
+}
+'''
+    
+    script_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "Models", "download-custom-model.ps1")
+    with open(script_path, "w") as f:
+        f.write(ps_script)
+    print(f"\n[OK] Created PowerShell fallback script: {script_path}")
+
+def main():
+    print("\n" + "="*60)
+    print("Custom Address Embedding Model - ONNX Converter")
+    print("="*60)
+    
+    # Check if we should just create the fallback script
+    if len(sys.argv) > 1 and sys.argv[1] == "--check-only":
+        check_requirements()
+        return 0
+    
+    # Check requirements first
+    print("Checking requirements...")
+    check_requirements()
+    
+    # Download and convert
+    if download_and_convert_model():
+        # Create fallback PowerShell script
+        create_powershell_download_script()
+        print("\n" + "="*60)
+        print("Conversion complete!")
+        print("="*60)
+        print("\nNext steps:")
+        print("1. Update appsettings.json to use the new model:")
+        print('   "Embedding": { "ModelName": "custom-all-MiniLM-L6-v2-address" }')
+        print("2. Update EmbeddingService.cs to support the new model path")
+        print("3. Run the application")
+        return 0
+    else:
+        return 1
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -93,17 +93,29 @@ public class EmbeddingService : IEmbeddingService
        var modelFileName = modelName switch
        {
            "sentence-transformers/all-MiniLM-L6-v2" => "all-MiniLM-L6-v2.onnx",
+            "jarredparrett/all-MiniLM-L6-v2_tuned_on_deepparse_address_mutations_comb_3" => "address-embedding-model.onnx",
+            "custom-all-MiniLM-L6-v2-address" => "address-embedding-model.onnx",
            _ => throw new NotSupportedException($"Model '{modelName}' is not supported for local inference")
        };

-        var modelPath = Path.Combine(AppContext.BaseDirectory, "Models", modelFileName);
-        
-        if (!File.Exists(modelPath))
+        // Check multiple possible locations for the model file
+        var possiblePaths = new[]
        {
-            modelPath = Path.Combine("Models", modelFileName);
+            Path.Combine(AppContext.BaseDirectory, "Models", modelFileName),
+            Path.Combine(AppContext.BaseDirectory, "Models", "Models", modelFileName),
+            Path.Combine("Models", modelFileName),
+            Path.Combine("Models", "Models", modelFileName)
+        };
+
+        foreach (var modelPath in possiblePaths)
+        {
+            if (File.Exists(modelPath))
+            {
+                return modelPath;
+            }
        }

-        return modelPath;
+        throw new FileNotFoundException($"Model file '{modelFileName}' not found. Searched in: {string.Join(", ", possiblePaths)}");
    }

    public Task<float[]> GenerateEmbeddingAsync(string text, CancellationToken cancellationToken = default)
@@ -21,6 +21,9 @@
    <None Update="Models\all-MiniLM-L6-v2.onnx">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
+    <None Update="Models\Models\address-embedding-model.onnx">
+      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
+    </None>
  </ItemGroup>

 </Project>
@@ -6,7 +6,7 @@
 		"CollectionName": "addresses"
 	},
 	"Embedding": {
-		"ModelName": "sentence-transformers/all-MiniLM-L6-v2",
+		"ModelName": "custom-all-MiniLM-L6-v2-address",
 		"Dimension": 384,
 		"ApiToken": "",
 		"UseLocalInference": true