Skip to content

Commit 2376523

Browse files
authored
Merge branch 'main' into semcaching
2 parents 7cf2156 + fe0b5b5 commit 2376523

File tree

12 files changed

+698
-303
lines changed

12 files changed

+698
-303
lines changed

config/config.yaml

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -128,7 +128,6 @@ classifier:
128128
threshold: 0.7
129129
use_cpu: true
130130
pii_mapping_path: "models/pii_classifier_modernbert-base_presidio_token_model/pii_type_mapping.json"
131-
load_aware: false
132131
categories:
133132
- name: business
134133
use_reasoning: false

deploy/kubernetes/config.yaml

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -78,7 +78,6 @@ classifier:
7878
threshold: 0.7
7979
use_cpu: true
8080
pii_mapping_path: "models/pii_classifier_modernbert-base_presidio_token_model/pii_type_mapping.json"
81-
load_aware: false
8281
categories:
8382
- name: business
8483
model_scores:

src/semantic-router/pkg/config/config.go

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -34,7 +34,6 @@ type RouterConfig struct {
3434
UseCPU bool `yaml:"use_cpu"`
3535
PIIMappingPath string `yaml:"pii_mapping_path"`
3636
} `yaml:"pii_model"`
37-
LoadAware bool `yaml:"load_aware"`
3837
} `yaml:"classifier"`
3938

4039
// Categories for routing queries

src/semantic-router/pkg/config/config_test.go

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -60,7 +60,6 @@ classifier:
6060
use_cpu: true
6161
use_modernbert: false
6262
pii_mapping_path: "/path/to/pii.json"
63-
load_aware: true
6463
6564
categories:
6665
- name: "general"
@@ -138,7 +137,6 @@ tools:
138137
// Verify classifier config
139138
Expect(cfg.Classifier.CategoryModel.ModelID).To(Equal("test-category-model"))
140139
Expect(cfg.Classifier.CategoryModel.UseModernBERT).To(BeTrue())
141-
Expect(cfg.Classifier.LoadAware).To(BeTrue())
142140

143141
// Verify categories
144142
Expect(cfg.Categories).To(HaveLen(1))

src/semantic-router/pkg/extproc/request_handler.go

Lines changed: 0 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -370,9 +370,6 @@ func (r *OpenAIRouter) handleModelRouting(openAIRequest *openai.ChatCompletionNe
370370
effortForMetrics := r.getReasoningEffort(categoryName)
371371
metrics.RecordReasoningDecision(categoryName, matchedModel, useReasoning, effortForMetrics)
372372

373-
// Track the model load for the selected model
374-
r.Classifier.IncrementModelLoad(matchedModel)
375-
376373
// Track the model routing change
377374
metrics.RecordModelRouting(originalModel, matchedModel)
378375

src/semantic-router/pkg/extproc/response_handler.go

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -52,7 +52,6 @@ func (r *OpenAIRouter) handleResponseBody(v *ext_proc.ProcessingRequest_Response
5252
float64(completionTokens),
5353
)
5454
metrics.RecordModelCompletionLatency(ctx.RequestModel, completionLatency.Seconds())
55-
r.Classifier.DecrementModelLoad(ctx.RequestModel)
5655

5756
// Compute and record cost if pricing is configured
5857
if r.Config != nil {

src/semantic-router/pkg/extproc/router.go

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -145,8 +145,7 @@ func NewOpenAIRouter(configPath string) (*OpenAIRouter, error) {
145145

146146
// Create utility components
147147
piiChecker := pii.NewPolicyChecker(cfg, cfg.ModelConfig)
148-
modelTTFT := make(map[string]float64) // Empty TTFT map since load balancing is disabled
149-
classifier := classification.NewClassifier(cfg, categoryMapping, piiMapping, jailbreakMapping, modelTTFT)
148+
classifier := classification.NewClassifier(cfg, categoryMapping, piiMapping, jailbreakMapping)
150149

151150
// Create global classification service for API access
152151
services.NewClassificationService(classifier, cfg)

src/semantic-router/pkg/extproc/security_test.go

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -52,7 +52,7 @@ var _ = Describe("Security Checks", func() {
5252
},
5353
}
5454
router.PIIChecker = pii.NewPolicyChecker(cfg, cfg.ModelConfig)
55-
router.Classifier = classification.NewClassifier(cfg, router.Classifier.CategoryMapping, router.Classifier.PIIMapping, nil, router.Classifier.ModelTTFT)
55+
router.Classifier = classification.NewClassifier(cfg, router.Classifier.CategoryMapping, router.Classifier.PIIMapping, nil)
5656
})
5757

5858
It("should allow requests with no PII", func() {
@@ -97,7 +97,7 @@ var _ = Describe("Security Checks", func() {
9797
piiMapping, err := classification.LoadPIIMapping(cfg.Classifier.PIIModel.PIIMappingPath)
9898
Expect(err).NotTo(HaveOccurred())
9999

100-
router.Classifier = classification.NewClassifier(cfg, router.Classifier.CategoryMapping, piiMapping, nil, router.Classifier.ModelTTFT)
100+
router.Classifier = classification.NewClassifier(cfg, router.Classifier.CategoryMapping, piiMapping, nil)
101101
})
102102

103103
Describe("ClassifyPII method", func() {
@@ -339,7 +339,7 @@ var _ = Describe("Security Checks", func() {
339339
piiMapping, err := classification.LoadPIIMapping(cfg.Classifier.PIIModel.PIIMappingPath)
340340
Expect(err).NotTo(HaveOccurred())
341341

342-
router.Classifier = classification.NewClassifier(cfg, router.Classifier.CategoryMapping, piiMapping, nil, router.Classifier.ModelTTFT)
342+
router.Classifier = classification.NewClassifier(cfg, router.Classifier.CategoryMapping, piiMapping, nil)
343343
})
344344

345345
Describe("Error handling and edge cases", func() {
@@ -524,7 +524,7 @@ var _ = Describe("Security Checks", func() {
524524
IdxToLabel: map[string]string{"0": "benign", "1": "jailbreak"},
525525
}
526526

527-
router.Classifier = classification.NewClassifier(cfg, router.Classifier.CategoryMapping, router.Classifier.PIIMapping, jailbreakMapping, router.Classifier.ModelTTFT)
527+
router.Classifier = classification.NewClassifier(cfg, router.Classifier.CategoryMapping, router.Classifier.PIIMapping, jailbreakMapping)
528528
})
529529

530530
It("should process potential jailbreak attempts", func() {

src/semantic-router/pkg/extproc/test_utils_test.go

Lines changed: 1 addition & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -95,7 +95,6 @@ func CreateTestConfig() *config.RouterConfig {
9595
UseCPU bool `yaml:"use_cpu"`
9696
PIIMappingPath string `yaml:"pii_mapping_path"`
9797
} `yaml:"pii_model"`
98-
LoadAware bool `yaml:"load_aware"`
9998
}{
10099
CategoryModel: struct {
101100
ModelID string `yaml:"model_id"`
@@ -119,7 +118,6 @@ func CreateTestConfig() *config.RouterConfig {
119118
UseCPU: true,
120119
PIIMappingPath: "../../../../models/pii_classifier_modernbert-base_presidio_token_model/pii_type_mapping.json",
121120
},
122-
LoadAware: true,
123121
},
124122
Categories: []config.Category{
125123
{
@@ -232,11 +230,7 @@ func CreateTestRouter(cfg *config.RouterConfig) (*extproc.OpenAIRouter, error) {
232230
toolsDatabase := tools.NewToolsDatabase(toolsOptions)
233231

234232
// Create classifier
235-
modelTTFT := map[string]float64{
236-
"model-a": 2.5,
237-
"model-b": 1.8,
238-
}
239-
classifier := classification.NewClassifier(cfg, categoryMapping, piiMapping, nil, modelTTFT)
233+
classifier := classification.NewClassifier(cfg, categoryMapping, piiMapping, nil)
240234

241235
// Create PII checker
242236
piiChecker := pii.NewPolicyChecker(cfg, cfg.ModelConfig)

0 commit comments

Comments
 (0)