fix(pii): enable LoRA PII auto-detection with minimal changes

Switch PII classification from hardcoded ModernBERT to auto-detecting
Candle BERT classifier. The Rust layer already has built-in auto-detection
that checks for lora_config.json and routes to LoRA or Traditional models.

Changes:
1. Init: Use InitCandleBertTokenClassifier (has auto-detect built-in)
2. Inference: Use ClassifyCandleBertTokens (auto-routes to initialized classifier)

This enables LoRA PII models to work automatically without config changes,
providing higher confidence scores for PII entity detection.

Fixes vllm-project#647

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>
Signed-off-by: Yossi Ovadia <[email protected]>

Author: yossiovadia

src/semantic-router/pkg/classification/classifier.go

@@ -139,35 +139,41 @@ func createJailbreakInference(useModernBERT bool) JailbreakInference {
 }
 
 type PIIInitializer interface {
-	Init(modelID string, useCPU bool) error
+	Init(modelID string, useCPU bool, numClasses int) error
 }
 
-type ModernBertPIIInitializer struct{}
+type PIIInitializerImpl struct{}
 
-func (c *ModernBertPIIInitializer) Init(modelID string, useCPU bool) error {
-	err := candle_binding.InitModernBertPIITokenClassifier(modelID, useCPU)
-	if err != nil {
-		return err
+func (c *PIIInitializerImpl) Init(modelID string, useCPU bool, numClasses int) error {
+	// Use auto-detecting Candle BERT init - checks for lora_config.json
+	success := candle_binding.InitCandleBertTokenClassifier(modelID, numClasses, useCPU)
+	if !success {
+		return fmt.Errorf("failed to initialize PII token classifier")
 	}
-	logging.Infof("Initialized ModernBERT PII token classifier for entity detection")
+	logging.Infof("Initialized PII token classifier with auto-detection (LoRA or Traditional)")
 	return nil
 }
 
-// createPIIInitializer creates the appropriate PII initializer (currently only ModernBERT)
-func createPIIInitializer() PIIInitializer { return &ModernBertPIIInitializer{} }
+// createPIIInitializer creates the PII initializer (auto-detecting)
+func createPIIInitializer() PIIInitializer {
+	return &PIIInitializerImpl{}
+}
 
 type PIIInference interface {
 	ClassifyTokens(text string, configPath string) (candle_binding.TokenClassificationResult, error)
 }
 
-type ModernBertPIIInference struct{}
+type PIIInferenceImpl struct{}
 
-func (c *ModernBertPIIInference) ClassifyTokens(text string, configPath string) (candle_binding.TokenClassificationResult, error) {
-	return candle_binding.ClassifyModernBertPIITokens(text, configPath)
+func (c *PIIInferenceImpl) ClassifyTokens(text string, configPath string) (candle_binding.TokenClassificationResult, error) {
+	// Auto-detecting inference - uses whichever classifier was initialized (LoRA or Traditional)
+	return candle_binding.ClassifyCandleBertTokens(text)
 }
 
-// createPIIInference creates the appropriate PII inference (currently only ModernBERT)
-func createPIIInference() PIIInference { return &ModernBertPIIInference{} }
+// createPIIInference creates the PII inference (auto-detecting)
+func createPIIInference() PIIInference {
+	return &PIIInferenceImpl{}
+}
 
 // JailbreakDetection represents the result of jailbreak analysis for a piece of content
 type JailbreakDetection struct {
@@ -350,7 +356,7 @@ func NewClassifier(cfg *config.RouterConfig, categoryMapping *CategoryMapping, p
 
 	// Add in-tree classifier if configured
 	if cfg.CategoryModel.ModelID != "" {
-		options = append(options, withCategory(categoryMapping, createCategoryInitializer(cfg.UseModernBERT), createCategoryInference(cfg.UseModernBERT)))
+		options = append(options, withCategory(categoryMapping, createCategoryInitializer(cfg.CategoryModel.UseModernBERT), createCategoryInference(cfg.CategoryModel.UseModernBERT)))
 	}
 
 	// Add MCP classifier if configured
@@ -511,7 +517,8 @@ func (c *Classifier) initializePIIClassifier() error {
 		return fmt.Errorf("not enough PII types for classification, need at least 2, got %d", numPIIClasses)
 	}
 
-	return c.piiInitializer.Init(c.Config.PIIModel.ModelID, c.Config.PIIModel.UseCPU)
+	// Pass numClasses to support auto-detection
+	return c.piiInitializer.Init(c.Config.PIIModel.ModelID, c.Config.PIIModel.UseCPU, numPIIClasses)
 }
 
 // ClassifyCategoryWithEntropy performs category classification with entropy-based reasoning decision