Add requirement inspection benchmarks

This commit is contained in:
2026-10-09 00:30:51 +00:00
parent 78bff91395
commit 04d5d695e0
2 changed files with 155 additions and 0 deletions

View File

@@ -15,6 +15,7 @@ const benchmarkDefinitionCount = 128
var (
constructionResult *loaded
selectedPrompt *domain.PromptDefinition
selectedPromptHash string
selectedProfile *domain.ExecutionProfile
)
@@ -177,3 +178,85 @@ func BenchmarkConfigurationSelection(b *testing.B) {
b.ReportMetric(float64(shape.effectiveEntries), "effective-entries/op")
})
}
func inspectionExtraParams(entries int) map[string]any {
providers := make(map[string]any, entries)
for index := range entries {
providers[fmt.Sprintf("provider-%03d", index)] = map[string]any{
"enabled": true,
"labels": []any{fmt.Sprintf("tier-%d", index%4), fmt.Sprintf("region-%d", index%3)},
"limits": map[string]any{"burst": index + 1, "period": "minute"},
"routes": []any{map[string]any{"weight": index + 1, "fallback": true}},
}
}
return map[string]any{"provider_options": providers}
}
func inspectionDefinitions(extraParams map[string]any) configurationShape {
profiles := map[string]any{
"small": map[string]any{"model": "inspection-model", "extra_params": extraParams},
}
prompt := benchmarkPrompt("small")
prompt["inputs"] = []any{map[string]any{"name": "transcript", "required": true}}
prompts := map[string]any{
"inspection": map[string]any{"1": prompt},
}
return configurationShape{
main: configkit.Input{Source: configkit.Mapping("inspection definitions", map[string]any{
"profiles": profiles,
"prompts": prompts,
})},
inputEntries: 2, effectiveEntries: 2, tiers: 1,
}
}
func BenchmarkInspectionSelection(b *testing.B) {
for _, scenario := range []struct {
name string
entries int
}{
{name: "small_extra", entries: 2},
{name: "nested_extra", entries: 64},
} {
shape := inspectionDefinitions(inspectionExtraParams(scenario.entries))
result, err := load(shape.main, nil)
if err != nil {
b.Fatal(err)
}
ctx := context.Background()
definition, hash, err := result.GetPromptSnapshot(ctx, "inspection", "1")
if err != nil || definition.ID != "inspection" || definition.Version != "1" || hash == "" ||
len(definition.Inputs) != 1 || definition.Inputs[0].Name != "transcript" || !definition.Inputs[0].Required {
b.Fatalf("prompt lookup fixture is invalid: definition=%+v hash=%q err=%v", definition, hash, err)
}
profile, err := result.GetProfile(ctx, "small")
if err != nil || profile.ID != "small" || profile.ResolvedTarget.Model != "inspection-model" ||
len(profile.ResolvedTarget.ExtraParams["provider_options"].(map[string]any)) != scenario.entries {
b.Fatalf("profile lookup fixture is invalid: profile=%+v err=%v", profile, err)
}
b.Run("prompt/"+scenario.name, func(b *testing.B) {
b.ReportAllocs()
for b.Loop() {
definition, hash, err := result.GetPromptSnapshot(ctx, "inspection", "1")
if err != nil {
b.Fatal(err)
}
selectedPrompt = definition
selectedPromptHash = hash
}
b.ReportMetric(float64(scenario.entries), "extra-entries/op")
})
b.Run("profile/"+scenario.name, func(b *testing.B) {
b.ReportAllocs()
for b.Loop() {
profile, err := result.GetProfile(ctx, "small")
if err != nil {
b.Fatal(err)
}
selectedProfile = profile
}
b.ReportMetric(float64(scenario.entries), "extra-entries/op")
})
}
}

View File

@@ -0,0 +1,72 @@
package promptkit_test
import (
"context"
"fmt"
"testing"
"gitea.maximumdirect.net/eric/ConfigKit"
"gitea.maximumdirect.net/eric/promptkit"
)
var inspectedRequirements *promptkit.RequirementInspection
func requirementInspectionExtras(entries int) map[string]any {
providers := make(map[string]any, entries)
for index := range entries {
providers[fmt.Sprintf("provider-%03d", index)] = map[string]any{
"enabled": true,
"labels": []any{fmt.Sprintf("tier-%d", index%4), fmt.Sprintf("region-%d", index%3)},
"limits": map[string]any{"burst": index + 1, "period": "minute"},
"routes": []any{map[string]any{"weight": index + 1, "fallback": true}},
}
}
return map[string]any{"provider_options": providers}
}
func BenchmarkRequirementInspection(b *testing.B) {
profiles := map[string]any{
"small": map[string]any{
"model": "inspection-model", "extra_params": requirementInspectionExtras(2),
},
"large": map[string]any{
"model": "inspection-model", "extra_params": requirementInspectionExtras(64),
},
}
definition := contractPrompt("small", "Hello {{.name}} {{input \"transcript\"}}", []any{
map[string]any{"name": "transcript", "required": true},
}, "", noValidationOutput)
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
Source: configkit.Mapping("requirement inspection benchmark", map[string]any{
"prompts": oneVersionPrompt("inspection", "1", definition),
"profiles": profiles,
}),
}}})
if err != nil {
b.Fatal(err)
}
ctx := context.Background()
for _, profileID := range []string{"small", "large"} {
request := promptkit.RequirementInspectionRequest{
PromptID: "inspection", ProfileID: profileID, Vars: map[string]string{"name": "Ada"},
}
inspection, err := engine.InspectRequirements(ctx, request)
if err != nil || inspection.PromptID != "inspection" || inspection.PromptVersion != "1" ||
inspection.ProfileID != profileID || !inspection.Complete || len(inspection.Inputs) != 1 ||
inspection.Inputs[0].Name != "transcript" || inspection.Inputs[0].Kind != promptkit.RequirementRequired ||
len(inspection.Variables) != 1 || inspection.Variables[0].Name != "name" || !inspection.Variables[0].Supplied {
b.Fatalf("requirement inspection fixture for %q is invalid: inspection=%+v err=%v", profileID, inspection, err)
}
selectedRequest := request
b.Run(profileID+"_extra", func(b *testing.B) {
b.ReportAllocs()
for b.Loop() {
inspection, err := engine.InspectRequirements(ctx, selectedRequest)
if err != nil {
b.Fatal(err)
}
inspectedRequirements = inspection
}
})
}
}