Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions packages/benchmark/src/benchmark.ts
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,19 @@ export interface JudgeModel {
) => Promise<TypedModelResponse<T>>;
}

export type TraceEvent =
| {phase: "user_message"; turn: number; durationMs: number}
| {phase: "assistant_response"; turn: number; durationMs: number}
| {phase: "judge"; slug: string; durationMs: number}
| {phase: "judges"; durationMs: number; judgeCount: number};

export interface TestContext {
getUserResponse: (request: ModelRequest) => Promise<ModelResponse>;
getAssistantResponse: (request: ModelRequest) => Promise<ModelResponse>;
/** Record of judge model slug → callable judge model. */
judgeModels: Record<string, JudgeModel>;
/** Optional observability hook. No-op when undefined. */
trace?: (event: TraceEvent) => void;
}

export interface GenerationEvent<T> {
Expand Down
27 changes: 24 additions & 3 deletions packages/benchmark/src/kora.ts
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@ import {
PinnedDemographics,
} from "./allocation/allocatePersonas.js";
import {makeRng, shuffleWith} from "./allocation/rng.js";
import {Benchmark, JudgeModel} from "./benchmark.js";
import {Benchmark, JudgeModel, TraceEvent} from "./benchmark.js";
import {
generateFirstUserMessage,
generateNextUserMessage,
Expand Down Expand Up @@ -78,7 +78,8 @@ export async function runJudges(
judgeModels: Record<string, JudgeModel>,
scenario: Scenario,
prompt: ScenarioPrompt,
messages: readonly ModelMessage[]
messages: readonly ModelMessage[],
trace?: (event: TraceEvent) => void
): Promise<TestResult> {
const riskCategory = RiskCategory.find(scenario.seed.riskCategoryId);
const risk = RiskCategory.findRisk(riskCategory, scenario.seed.riskId);
Expand All @@ -94,10 +95,12 @@ export async function runJudges(
);

const judgeSlugs = Object.keys(judgeModels);
const tJudges = Date.now();
const judgeAssessments = await Promise.all(
judgeSlugs.map(async (slug): Promise<JudgeAssessment> => {
const judge = judgeModels[slug]!;

const tJudge = Date.now();
const [safetyResult, mechanismResult] = await Promise.all([
judge.getResponse({
messages: [
Expand All @@ -114,6 +117,7 @@ export async function runJudges(
outputType: MechanismAssessment.io,
}),
]);
trace?.({phase: "judge", slug, durationMs: Date.now() - tJudge});

return {
judgeModelSlug: slug,
Expand All @@ -122,6 +126,11 @@ export async function runJudges(
};
})
);
trace?.({
phase: "judges",
durationMs: Date.now() - tJudges,
judgeCount: judgeSlugs.length,
});

const assessment = aggregateTestAssessments(
judgeAssessments.map(j => j.assessment)
Expand Down Expand Up @@ -448,13 +457,19 @@ export const kora = Benchmark.new({
const startTurn = messages.length / 2;

for (let i = startTurn; i < conversationLength; i++) {
const tUserMessage = Date.now();
const userMessage = await (() => {
if (i === 0) {
return scenario.firstUserMessage;
}

return generateNextUserMessage(c, risk, scenario, messages);
})();
c.trace?.({
phase: "user_message",
turn: i,
durationMs: Date.now() - tUserMessage,
});

// TODO: Maybe assert refusal of user model.

Expand All @@ -463,6 +478,7 @@ export const kora = Benchmark.new({
content: userMessage,
});

const tAssistant = Date.now();
const modelMessage = await (async () => {
const modelPrompt = conversationToNextMessagePrompt({
ageRange: promptAgeRange,
Expand All @@ -479,14 +495,19 @@ export const kora = Benchmark.new({
});
return output;
})();
c.trace?.({
phase: "assistant_response",
turn: i,
durationMs: Date.now() - tAssistant,
});

messages.push({
role: "assistant",
content: modelMessage,
});
}

return runJudges(c.judgeModels, scenario, prompt, messages);
return runJudges(c.judgeModels, scenario, prompt, messages, c.trace);
},
mapTestResultToRunResult(result) {
const {assessment, mechanismAssessment} = result;
Expand Down
Loading