oh-my-knowledge 0.0.1 → 0.19.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/LICENSE +21 -0
- package/README.md +600 -322
- package/README.zh.md +724 -0
- package/dist/src/analysis/coverage-analyzer.d.ts +57 -0
- package/dist/src/analysis/coverage-analyzer.d.ts.map +1 -0
- package/dist/src/analysis/coverage-analyzer.js +262 -0
- package/dist/src/analysis/coverage-analyzer.js.map +1 -0
- package/dist/src/analysis/gap-analyzer.d.ts +114 -0
- package/dist/src/analysis/gap-analyzer.d.ts.map +1 -0
- package/dist/src/analysis/gap-analyzer.js +439 -0
- package/dist/src/analysis/gap-analyzer.js.map +1 -0
- package/dist/src/analysis/hedging-classifier.d.ts +34 -0
- package/dist/src/analysis/hedging-classifier.d.ts.map +1 -0
- package/dist/src/analysis/hedging-classifier.js +144 -0
- package/dist/src/analysis/hedging-classifier.js.map +1 -0
- package/dist/src/analysis/report-diagnostics.d.ts +9 -0
- package/dist/src/analysis/report-diagnostics.d.ts.map +1 -0
- package/dist/src/analysis/report-diagnostics.js +631 -0
- package/dist/src/analysis/report-diagnostics.js.map +1 -0
- package/dist/src/authoring/evolver.d.ts +64 -0
- package/dist/src/authoring/evolver.d.ts.map +1 -0
- package/dist/src/authoring/evolver.js +275 -0
- package/dist/src/authoring/evolver.js.map +1 -0
- package/dist/src/authoring/generator.d.ts +13 -0
- package/dist/src/authoring/generator.d.ts.map +1 -0
- package/dist/src/authoring/generator.js +60 -0
- package/dist/src/authoring/generator.js.map +1 -0
- package/dist/src/cli.d.ts +3 -0
- package/dist/src/cli.d.ts.map +1 -0
- package/dist/src/cli.js +1012 -0
- package/dist/src/cli.js.map +1 -0
- package/dist/src/eval-core/cache.d.ts +11 -0
- package/dist/src/eval-core/cache.d.ts.map +1 -0
- package/dist/src/eval-core/cache.js +53 -0
- package/dist/src/eval-core/cache.js.map +1 -0
- package/dist/src/eval-core/ci-gates.d.ts +17 -0
- package/dist/src/eval-core/ci-gates.d.ts.map +1 -0
- package/dist/src/eval-core/ci-gates.js +42 -0
- package/dist/src/eval-core/ci-gates.js.map +1 -0
- package/dist/src/eval-core/dependency-checker.d.ts +37 -0
- package/dist/src/eval-core/dependency-checker.d.ts.map +1 -0
- package/dist/src/eval-core/dependency-checker.js +280 -0
- package/dist/src/eval-core/dependency-checker.js.map +1 -0
- package/dist/src/eval-core/evaluation-execution.d.ts +26 -0
- package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -0
- package/dist/src/eval-core/evaluation-execution.js +196 -0
- package/dist/src/eval-core/evaluation-execution.js.map +1 -0
- package/dist/src/eval-core/evaluation-job.d.ts +50 -0
- package/dist/src/eval-core/evaluation-job.d.ts.map +1 -0
- package/dist/src/eval-core/evaluation-job.js +114 -0
- package/dist/src/eval-core/evaluation-job.js.map +1 -0
- package/dist/src/eval-core/evaluation-reporting.d.ts +28 -0
- package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -0
- package/dist/src/eval-core/evaluation-reporting.js +123 -0
- package/dist/src/eval-core/evaluation-reporting.js.map +1 -0
- package/dist/src/eval-core/execution-strategy.d.ts +11 -0
- package/dist/src/eval-core/execution-strategy.d.ts.map +1 -0
- package/dist/src/eval-core/execution-strategy.js +121 -0
- package/dist/src/eval-core/execution-strategy.js.map +1 -0
- package/dist/src/eval-core/fact-checker.d.ts +25 -0
- package/dist/src/eval-core/fact-checker.d.ts.map +1 -0
- package/dist/src/eval-core/fact-checker.js +66 -0
- package/dist/src/eval-core/fact-checker.js.map +1 -0
- package/dist/src/eval-core/schema.d.ts +30 -0
- package/dist/src/eval-core/schema.d.ts.map +1 -0
- package/dist/src/eval-core/schema.js +195 -0
- package/dist/src/eval-core/schema.js.map +1 -0
- package/dist/src/eval-core/statistics.d.ts +86 -0
- package/dist/src/eval-core/statistics.d.ts.map +1 -0
- package/dist/src/eval-core/statistics.js +210 -0
- package/dist/src/eval-core/statistics.js.map +1 -0
- package/dist/src/eval-core/task-planner.d.ts +4 -0
- package/dist/src/eval-core/task-planner.d.ts.map +1 -0
- package/dist/src/eval-core/task-planner.js +33 -0
- package/dist/src/eval-core/task-planner.js.map +1 -0
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +137 -0
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +1 -0
- package/dist/src/eval-workflows/each-evaluation-workflow.js +174 -0
- package/dist/src/eval-workflows/each-evaluation-workflow.js.map +1 -0
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts +42 -0
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -0
- package/dist/src/eval-workflows/evaluation-pipeline.js +218 -0
- package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -0
- package/dist/src/eval-workflows/evaluation-preparation.d.ts +64 -0
- package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -0
- package/dist/src/eval-workflows/evaluation-preparation.js +90 -0
- package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -0
- package/dist/src/eval-workflows/run-evaluation.d.ts +111 -0
- package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -0
- package/dist/src/eval-workflows/run-evaluation.js +272 -0
- package/dist/src/eval-workflows/run-evaluation.js.map +1 -0
- package/dist/src/executors/anthropic-api.d.ts +3 -0
- package/dist/src/executors/anthropic-api.d.ts.map +1 -0
- package/dist/src/executors/anthropic-api.js +43 -0
- package/dist/src/executors/anthropic-api.js.map +1 -0
- package/dist/src/executors/claude-cli.d.ts +3 -0
- package/dist/src/executors/claude-cli.d.ts.map +1 -0
- package/dist/src/executors/claude-cli.js +103 -0
- package/dist/src/executors/claude-cli.js.map +1 -0
- package/dist/src/executors/claude-sdk-trace.d.ts +10 -0
- package/dist/src/executors/claude-sdk-trace.d.ts.map +1 -0
- package/dist/src/executors/claude-sdk-trace.js +100 -0
- package/dist/src/executors/claude-sdk-trace.js.map +1 -0
- package/dist/src/executors/claude-sdk.d.ts +3 -0
- package/dist/src/executors/claude-sdk.d.ts.map +1 -0
- package/dist/src/executors/claude-sdk.js +160 -0
- package/dist/src/executors/claude-sdk.js.map +1 -0
- package/dist/src/executors/gemini.d.ts +3 -0
- package/dist/src/executors/gemini.d.ts.map +1 -0
- package/dist/src/executors/gemini.js +78 -0
- package/dist/src/executors/gemini.js.map +1 -0
- package/dist/src/executors/index.d.ts +7 -0
- package/dist/src/executors/index.d.ts.map +1 -0
- package/dist/src/executors/index.js +22 -0
- package/dist/src/executors/index.js.map +1 -0
- package/dist/src/executors/openai-api.d.ts +3 -0
- package/dist/src/executors/openai-api.d.ts.map +1 -0
- package/dist/src/executors/openai-api.js +40 -0
- package/dist/src/executors/openai-api.js.map +1 -0
- package/dist/src/executors/openai-cli.d.ts +3 -0
- package/dist/src/executors/openai-cli.d.ts.map +1 -0
- package/dist/src/executors/openai-cli.js +60 -0
- package/dist/src/executors/openai-cli.js.map +1 -0
- package/dist/src/executors/script.d.ts +3 -0
- package/dist/src/executors/script.d.ts.map +1 -0
- package/dist/src/executors/script.js +63 -0
- package/dist/src/executors/script.js.map +1 -0
- package/dist/src/executors/shared.d.ts +117 -0
- package/dist/src/executors/shared.d.ts.map +1 -0
- package/dist/src/executors/shared.js +49 -0
- package/dist/src/executors/shared.js.map +1 -0
- package/dist/src/grading/assertions.d.ts +18 -0
- package/dist/src/grading/assertions.d.ts.map +1 -0
- package/dist/src/grading/assertions.js +239 -0
- package/dist/src/grading/assertions.js.map +1 -0
- package/dist/src/grading/index.d.ts +26 -0
- package/dist/src/grading/index.d.ts.map +1 -0
- package/dist/src/grading/index.js +98 -0
- package/dist/src/grading/index.js.map +1 -0
- package/dist/src/grading/judge.d.ts +13 -0
- package/dist/src/grading/judge.d.ts.map +1 -0
- package/dist/src/grading/judge.js +98 -0
- package/dist/src/grading/judge.js.map +1 -0
- package/dist/src/grading/layered-scores.d.ts +13 -0
- package/dist/src/grading/layered-scores.d.ts.map +1 -0
- package/dist/src/grading/layered-scores.js +62 -0
- package/dist/src/grading/layered-scores.js.map +1 -0
- package/dist/src/inputs/eval-config.d.ts +13 -0
- package/dist/src/inputs/eval-config.d.ts.map +1 -0
- package/dist/src/inputs/eval-config.js +136 -0
- package/dist/src/inputs/eval-config.js.map +1 -0
- package/dist/src/inputs/load-samples.d.ts +16 -0
- package/dist/src/inputs/load-samples.d.ts.map +1 -0
- package/dist/src/inputs/load-samples.js +51 -0
- package/dist/src/inputs/load-samples.js.map +1 -0
- package/dist/src/inputs/mcp-resolver.d.ts +50 -0
- package/dist/src/inputs/mcp-resolver.d.ts.map +1 -0
- package/dist/src/inputs/mcp-resolver.js +307 -0
- package/dist/src/inputs/mcp-resolver.js.map +1 -0
- package/dist/src/inputs/skill-loader.d.ts +18 -0
- package/dist/src/inputs/skill-loader.d.ts.map +1 -0
- package/dist/src/inputs/skill-loader.js +222 -0
- package/dist/src/inputs/skill-loader.js.map +1 -0
- package/dist/src/inputs/url-fetcher.d.ts +19 -0
- package/dist/src/inputs/url-fetcher.d.ts.map +1 -0
- package/dist/src/inputs/url-fetcher.js +241 -0
- package/dist/src/inputs/url-fetcher.js.map +1 -0
- package/dist/src/observability/skill-health-analyzer.d.ts +83 -0
- package/dist/src/observability/skill-health-analyzer.d.ts.map +1 -0
- package/dist/src/observability/skill-health-analyzer.js +238 -0
- package/dist/src/observability/skill-health-analyzer.js.map +1 -0
- package/dist/src/observability/trace-adapter.d.ts +75 -0
- package/dist/src/observability/trace-adapter.d.ts.map +1 -0
- package/dist/src/observability/trace-adapter.js +367 -0
- package/dist/src/observability/trace-adapter.js.map +1 -0
- package/dist/src/renderer/html-renderer.d.ts +9 -0
- package/dist/src/renderer/html-renderer.d.ts.map +1 -0
- package/dist/src/renderer/html-renderer.js +360 -0
- package/dist/src/renderer/html-renderer.js.map +1 -0
- package/dist/src/renderer/layout.d.ts +13 -0
- package/dist/src/renderer/layout.d.ts.map +1 -0
- package/dist/src/renderer/layout.js +582 -0
- package/dist/src/renderer/layout.js.map +1 -0
- package/dist/src/renderer/skill-health-renderer.d.ts +20 -0
- package/dist/src/renderer/skill-health-renderer.d.ts.map +1 -0
- package/dist/src/renderer/skill-health-renderer.js +253 -0
- package/dist/src/renderer/skill-health-renderer.js.map +1 -0
- package/dist/src/renderer/summary.d.ts +21 -0
- package/dist/src/renderer/summary.d.ts.map +1 -0
- package/dist/src/renderer/summary.js +1072 -0
- package/dist/src/renderer/summary.js.map +1 -0
- package/dist/src/renderer/table.d.ts +3 -0
- package/dist/src/renderer/table.d.ts.map +1 -0
- package/dist/src/renderer/table.js +134 -0
- package/dist/src/renderer/table.js.map +1 -0
- package/dist/src/renderer/trends.d.ts +6 -0
- package/dist/src/renderer/trends.d.ts.map +1 -0
- package/dist/src/renderer/trends.js +127 -0
- package/dist/src/renderer/trends.js.map +1 -0
- package/dist/src/server/job-store.d.ts +4 -0
- package/dist/src/server/job-store.d.ts.map +1 -0
- package/dist/src/server/job-store.js +68 -0
- package/dist/src/server/job-store.js.map +1 -0
- package/dist/src/server/report-server.d.ts +17 -0
- package/dist/src/server/report-server.d.ts.map +1 -0
- package/dist/src/server/report-server.js +591 -0
- package/dist/src/server/report-server.js.map +1 -0
- package/dist/src/server/report-store.d.ts +44 -0
- package/dist/src/server/report-store.d.ts.map +1 -0
- package/dist/src/server/report-store.js +190 -0
- package/dist/src/server/report-store.js.map +1 -0
- package/dist/src/types.d.ts +559 -0
- package/dist/src/types.d.ts.map +1 -0
- package/dist/src/types.js +2 -0
- package/dist/src/types.js.map +1 -0
- package/package.json +53 -5
|
@@ -0,0 +1,218 @@
|
|
|
1
|
+
import { createHash } from 'node:crypto';
|
|
2
|
+
import { existsSync, readFileSync } from 'node:fs';
|
|
3
|
+
import { analyzeResults } from '../analysis/report-diagnostics.js';
|
|
4
|
+
import { computeReportCoverage } from '../analysis/coverage-analyzer.js';
|
|
5
|
+
import { computeReportGapRates } from '../analysis/gap-analyzer.js';
|
|
6
|
+
import { aggregateReport, applyBlindMode, DEFAULT_OUTPUT_DIR, generateRunId, persistReport } from '../eval-core/evaluation-reporting.js';
|
|
7
|
+
import { executeTasks, preflight } from '../eval-core/evaluation-execution.js';
|
|
8
|
+
import { preflightDependencies, formatDependencyErrors } from '../eval-core/dependency-checker.js';
|
|
9
|
+
import { createFileJobStore, DEFAULT_JOBS_DIR, } from '../server/job-store.js';
|
|
10
|
+
import { stopAllServers } from '../inputs/mcp-resolver.js';
|
|
11
|
+
import { buildEvaluationRequest, createFailedJob, createEvaluationRun, createQueuedJob, createSucceededJob, finalizeEvaluationRun, markJobRunning, failEvaluationRun, } from '../eval-core/evaluation-job.js';
|
|
12
|
+
async function initializeEvaluationRunState({ samplesPath, skillDir, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, concurrency, timeoutMs, noCache, blind, project, owner, tags, runId, jobStore, persistJob, repeat, each, }) {
|
|
13
|
+
const request = buildEvaluationRequest({
|
|
14
|
+
samplesPath,
|
|
15
|
+
skillDir,
|
|
16
|
+
artifacts,
|
|
17
|
+
model,
|
|
18
|
+
judgeModel: noJudge ? null : judgeModel,
|
|
19
|
+
executor: executorName,
|
|
20
|
+
judgeExecutor: judgeExecutorName,
|
|
21
|
+
noJudge,
|
|
22
|
+
concurrency,
|
|
23
|
+
timeoutMs,
|
|
24
|
+
noCache,
|
|
25
|
+
dryRun: false,
|
|
26
|
+
blind,
|
|
27
|
+
project,
|
|
28
|
+
owner,
|
|
29
|
+
tags,
|
|
30
|
+
repeat,
|
|
31
|
+
each,
|
|
32
|
+
});
|
|
33
|
+
const createdAt = new Date().toISOString();
|
|
34
|
+
const { run: initialRun, startedAt } = createEvaluationRun(runId, createdAt);
|
|
35
|
+
const jobId = `job-${runId}`;
|
|
36
|
+
const resolvedJobStore = persistJob ? (jobStore ?? createFileJobStore(DEFAULT_JOBS_DIR)) : null;
|
|
37
|
+
const queuedJob = createQueuedJob({ jobId, request, createdAt });
|
|
38
|
+
if (resolvedJobStore)
|
|
39
|
+
await resolvedJobStore.save(jobId, queuedJob);
|
|
40
|
+
const runningJob = markJobRunning(queuedJob, runId, startedAt);
|
|
41
|
+
if (resolvedJobStore)
|
|
42
|
+
await resolvedJobStore.save(jobId, runningJob);
|
|
43
|
+
return { request, runId, jobId, createdAt, startedAt, initialRun, runningJob, resolvedJobStore };
|
|
44
|
+
}
|
|
45
|
+
function finalizeSuccessfulRun(state) {
|
|
46
|
+
const finishedAt = new Date().toISOString();
|
|
47
|
+
const run = finalizeEvaluationRun(state.initialRun, finishedAt);
|
|
48
|
+
const job = createSucceededJob({
|
|
49
|
+
jobId: state.jobId,
|
|
50
|
+
runId: state.runId,
|
|
51
|
+
reportId: state.runId,
|
|
52
|
+
request: state.request,
|
|
53
|
+
createdAt: state.createdAt,
|
|
54
|
+
startedAt: state.startedAt,
|
|
55
|
+
finishedAt,
|
|
56
|
+
});
|
|
57
|
+
return { run, job };
|
|
58
|
+
}
|
|
59
|
+
async function persistSuccessfulJob(state, job) {
|
|
60
|
+
if (state.resolvedJobStore) {
|
|
61
|
+
await state.resolvedJobStore.save(state.jobId, job);
|
|
62
|
+
}
|
|
63
|
+
}
|
|
64
|
+
async function persistFailedJob(state, err) {
|
|
65
|
+
const finishedAt = new Date().toISOString();
|
|
66
|
+
const failedJob = createFailedJob({
|
|
67
|
+
job: { ...state.runningJob, runId: state.runId, startedAt: state.startedAt, finishedAt: undefined },
|
|
68
|
+
error: err instanceof Error ? err.message : String(err),
|
|
69
|
+
finishedAt,
|
|
70
|
+
});
|
|
71
|
+
void failEvaluationRun(state.initialRun, finishedAt);
|
|
72
|
+
if (state.resolvedJobStore) {
|
|
73
|
+
await state.resolvedJobStore.save(state.jobId, failedJob);
|
|
74
|
+
}
|
|
75
|
+
}
|
|
76
|
+
/**
|
|
77
|
+
* Compute the mandatory test set watermark hash (spec §7.1). Returns the first
|
|
78
|
+
* 12 hex chars of SHA-256 over the samples file contents, or null if the file
|
|
79
|
+
* can't be read.
|
|
80
|
+
*/
|
|
81
|
+
function computeTestSetHash(samplesPath) {
|
|
82
|
+
if (!samplesPath || !existsSync(samplesPath))
|
|
83
|
+
return null;
|
|
84
|
+
try {
|
|
85
|
+
const content = readFileSync(samplesPath, 'utf-8');
|
|
86
|
+
return createHash('sha256').update(content).digest('hex').slice(0, 12);
|
|
87
|
+
}
|
|
88
|
+
catch {
|
|
89
|
+
return null;
|
|
90
|
+
}
|
|
91
|
+
}
|
|
92
|
+
function finalizeEvaluationReport({ report, results, artifacts, variantNames, blind, samplesPath, }) {
|
|
93
|
+
report.analysis = analyzeResults(report);
|
|
94
|
+
const hasToolData = Object.values(results).some((sampleResults) => (Object.values(sampleResults).some((variantResult) => variantResult.toolCalls && variantResult.toolCalls.length > 0)));
|
|
95
|
+
if (hasToolData) {
|
|
96
|
+
const artifactContents = Object.fromEntries(artifacts.map((artifact) => [artifact.name, artifact.content]));
|
|
97
|
+
const artifactCwds = Object.fromEntries(artifacts.map((artifact) => [artifact.name, artifact.cwd || null]));
|
|
98
|
+
const coverage = computeReportCoverage(report, artifactContents, artifactCwds);
|
|
99
|
+
if (Object.keys(coverage).length > 0) {
|
|
100
|
+
report.analysis.coverage = coverage;
|
|
101
|
+
}
|
|
102
|
+
}
|
|
103
|
+
// Gap rate computation runs on every successful report regardless of whether
|
|
104
|
+
// tool trace data is present — text-based signals (markers, hedging) still
|
|
105
|
+
// apply. The samples-file SHA is the mandatory watermark required by spec §7.1.
|
|
106
|
+
const gapReports = computeReportGapRates(report.results, variantNames);
|
|
107
|
+
if (Object.keys(gapReports).length > 0) {
|
|
108
|
+
const testSetHash = computeTestSetHash(samplesPath);
|
|
109
|
+
for (const variant of variantNames) {
|
|
110
|
+
const gr = gapReports[variant];
|
|
111
|
+
if (!gr)
|
|
112
|
+
continue;
|
|
113
|
+
gr.testSetPath = samplesPath;
|
|
114
|
+
gr.testSetHash = testSetHash;
|
|
115
|
+
}
|
|
116
|
+
report.analysis.gapReports = gapReports;
|
|
117
|
+
}
|
|
118
|
+
if (blind) {
|
|
119
|
+
applyBlindMode(report, variantNames, `${variantNames.join(',')}:${samplesPath}`);
|
|
120
|
+
}
|
|
121
|
+
return report;
|
|
122
|
+
}
|
|
123
|
+
export async function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, blind = false, concurrency = 1, timeoutMs, noCache = false, jobStore = null, persistJob = true, onProgress = null, skipPreflight = false, verbose = false, retry = 0, existingResults, requires, layeredStats = false, repeat, each, }) {
|
|
124
|
+
const variantNames = artifacts.map((artifact) => artifact.name);
|
|
125
|
+
const runState = await initializeEvaluationRunState({
|
|
126
|
+
samplesPath,
|
|
127
|
+
skillDir,
|
|
128
|
+
artifacts,
|
|
129
|
+
model,
|
|
130
|
+
judgeModel,
|
|
131
|
+
noJudge,
|
|
132
|
+
executorName,
|
|
133
|
+
judgeExecutorName,
|
|
134
|
+
concurrency,
|
|
135
|
+
timeoutMs,
|
|
136
|
+
noCache,
|
|
137
|
+
blind,
|
|
138
|
+
project,
|
|
139
|
+
owner,
|
|
140
|
+
tags,
|
|
141
|
+
runId: generateRunId(variantNames),
|
|
142
|
+
jobStore,
|
|
143
|
+
persistJob,
|
|
144
|
+
repeat,
|
|
145
|
+
each,
|
|
146
|
+
});
|
|
147
|
+
try {
|
|
148
|
+
if (!skipPreflight) {
|
|
149
|
+
if (onProgress)
|
|
150
|
+
onProgress({ phase: 'preflight', jobId: runState.jobId });
|
|
151
|
+
await preflight(executor, model);
|
|
152
|
+
if (!noJudge)
|
|
153
|
+
await preflight(judgeExecutor, judgeModel);
|
|
154
|
+
// Dependency check: auto-extract from skill contents + merge explicit requires
|
|
155
|
+
const skillContents = artifacts.map((a) => a.content).filter((c) => typeof c === 'string');
|
|
156
|
+
const cwd = artifacts.find((a) => a.cwd)?.cwd || skillDir || process.cwd();
|
|
157
|
+
const depResult = await preflightDependencies(skillContents, samples, cwd, requires, artifacts);
|
|
158
|
+
if (!depResult.ok) {
|
|
159
|
+
throw new Error(formatDependencyErrors(depResult.missing));
|
|
160
|
+
}
|
|
161
|
+
}
|
|
162
|
+
const { results, totalCostUSD, skipped } = await executeTasks({
|
|
163
|
+
tasks,
|
|
164
|
+
executor,
|
|
165
|
+
judgeExecutor,
|
|
166
|
+
model,
|
|
167
|
+
judgeModel,
|
|
168
|
+
noJudge,
|
|
169
|
+
samplesPath,
|
|
170
|
+
concurrency,
|
|
171
|
+
timeoutMs,
|
|
172
|
+
noCache,
|
|
173
|
+
verbose,
|
|
174
|
+
onProgress,
|
|
175
|
+
retry,
|
|
176
|
+
existingResults,
|
|
177
|
+
});
|
|
178
|
+
if (skipped > 0 && onProgress) {
|
|
179
|
+
onProgress({ phase: 'done', completed: tasks.length, total: tasks.length, sample_id: '', variant: '', skipped: true });
|
|
180
|
+
}
|
|
181
|
+
const { run, job } = finalizeSuccessfulRun(runState);
|
|
182
|
+
const report = finalizeEvaluationReport({
|
|
183
|
+
report: aggregateReport({
|
|
184
|
+
runId: runState.runId,
|
|
185
|
+
variants: variantNames,
|
|
186
|
+
model,
|
|
187
|
+
judgeModel,
|
|
188
|
+
noJudge,
|
|
189
|
+
executorName,
|
|
190
|
+
samples,
|
|
191
|
+
tasks,
|
|
192
|
+
results,
|
|
193
|
+
totalCostUSD,
|
|
194
|
+
artifacts,
|
|
195
|
+
request: runState.request,
|
|
196
|
+
run,
|
|
197
|
+
job,
|
|
198
|
+
layeredStats,
|
|
199
|
+
}),
|
|
200
|
+
results,
|
|
201
|
+
artifacts,
|
|
202
|
+
variantNames,
|
|
203
|
+
blind,
|
|
204
|
+
samplesPath,
|
|
205
|
+
});
|
|
206
|
+
const filePath = persistReport(report, outputDir);
|
|
207
|
+
await persistSuccessfulJob(runState, job);
|
|
208
|
+
return { report, filePath };
|
|
209
|
+
}
|
|
210
|
+
catch (err) {
|
|
211
|
+
await persistFailedJob(runState, err);
|
|
212
|
+
throw err;
|
|
213
|
+
}
|
|
214
|
+
finally {
|
|
215
|
+
await stopAllServers();
|
|
216
|
+
}
|
|
217
|
+
}
|
|
218
|
+
//# sourceMappingURL=evaluation-pipeline.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-pipeline.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,cAAc,EAAE,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,cAAc,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzI,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,sCAAsC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAEnG,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3D,OAAO,EACL,sBAAsB,EACtB,eAAe,EACf,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,iBAAiB,GAClB,MAAM,gCAAgC,CAAC;AA4BxC,KAAK,UAAU,4BAA4B,CAAC,EAC1C,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,WAAW,EACX,SAAS,EACT,OAAO,EACP,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,IAAI,GAsBL;IACC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB;QAChC,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,IAAI;KACL,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,KAAK,GAAG,OAAO,KAAK,EAAE,CAAC;IAC7B,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,MAAM,SAAS,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;IACjE,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IACpE,MAAM,UAAU,GAAG,cAAc,CAAC,SAAS,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;IAC/D,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,UAAU,CAAC,CAAC;IACrE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,UAAU,EAAE,UAAU,EAAE,gBAAgB,EAAE,CAAC;AACnG,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAyB;IACtD,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAChE,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,QAAQ,EAAE,KAAK,CAAC,KAAK;QACrB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,UAAU;KACX,CAAC,CAAC;IACH,OAAO,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;AACtB,CAAC;AAED,KAAK,UAAU,oBAAoB,CAAC,KAAyB,EAAE,GAAkB;IAC/E,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IACtD,CAAC;AACH,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,KAAyB,EAAE,GAAY;IACrE,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,SAAS,GAAG,eAAe,CAAC;QAChC,GAAG,EAAE,EAAE,GAAG,KAAK,CAAC,UAAU,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE;QACnG,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;QACvD,UAAU;KACX,CAAC,CAAC;IACH,KAAK,iBAAiB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IACrD,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC5D,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,SAAS,kBAAkB,CAAC,WAAmB;IAC7C,IAAI,CAAC,WAAW,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,IAAI,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QACnD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACzE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,wBAAwB,CAAC,EAChC,MAAM,EACN,OAAO,EACP,SAAS,EACT,YAAY,EACZ,KAAK,EACL,WAAW,GAQZ;IACC,MAAM,CAAC,QAAQ,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;IAEzC,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,CACjE,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,aAAa,CAAC,SAAS,IAAI,aAAa,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CACpH,CAAC,CAAC;IACH,IAAI,WAAW,EAAE,CAAC;QAChB,MAAM,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,EAAE,gBAAgB,EAAE,YAAY,CAAC,CAAC;QAC/E,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACrC,MAAM,CAAC,QAAS,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACvC,CAAC;IACH,CAAC;IAED,6EAA6E;IAC7E,2EAA2E;IAC3E,gFAAgF;IAChF,MAAM,UAAU,GAAG,qBAAqB,CAAC,MAAM,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;IACvE,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvC,MAAM,WAAW,GAAG,kBAAkB,CAAC,WAAW,CAAC,CAAC;QACpD,KAAK,MAAM,OAAO,IAAI,YAAY,EAAE,CAAC;YACnC,MAAM,EAAE,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC;YAC/B,IAAI,CAAC,EAAE;gBAAE,SAAS;YAClB,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;YAC7B,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,CAAC;QACD,MAAM,CAAC,QAAS,CAAC,UAAU,GAAG,UAAU,CAAC;IAC3C,CAAC;IAED,IAAI,KAAK,EAAE,CAAC;QACV,cAAc,CAAC,MAAM,EAAE,YAAY,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,WAAW,EAAE,CAAC,CAAC;IACnF,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAsCD,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,aAAa,GAAG,KAAK,EACrB,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,eAAe,EACf,QAAQ,EACR,YAAY,GAAG,KAAK,EACpB,MAAM,EACN,IAAI,GACsB;IAC1B,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,4BAA4B,CAAC;QAClD,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,WAAW;QACX,SAAS;QACT,OAAO;QACP,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK,EAAE,aAAa,CAAC,YAAY,CAAC;QAClC,QAAQ;QACR,UAAU;QACV,MAAM;QACN,IAAI;KACL,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,IAAI,CAAC,aAAa,EAAE,CAAC;YACnB,IAAI,UAAU;gBAAE,UAAU,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;YAC1E,MAAM,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YACjC,IAAI,CAAC,OAAO;gBAAE,MAAM,SAAS,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;YAEzD,+EAA+E;YAC/E,MAAM,aAAa,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;YACxG,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,GAAG,IAAI,QAAQ,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAC3E,MAAM,SAAS,GAAG,MAAM,qBAAqB,CAAC,aAAa,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC;YAChG,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,CAAC;gBAClB,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;YAC7D,CAAC;QACH,CAAC;QAED,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,GAAG,MAAM,YAAY,CAAC;YAC5D,KAAK;YACL,QAAQ;YACR,aAAa;YACb,KAAK;YACL,UAAU;YACV,OAAO;YACP,WAAW;YACX,WAAW;YACX,SAAS;YACT,OAAO;YACP,OAAO;YACP,UAAU;YACV,KAAK;YACL,eAAe;SAChB,CAAC,CAAC;QACH,IAAI,OAAO,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC;YAC9B,UAAU,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzH,CAAC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,wBAAwB,CAAC;YACtC,MAAM,EAAE,eAAe,CAAC;gBACtB,KAAK,EAAE,QAAQ,CAAC,KAAK;gBACrB,QAAQ,EAAE,YAAY;gBACtB,KAAK;gBACL,UAAU;gBACV,OAAO;gBACP,YAAY;gBACZ,OAAO;gBACP,KAAK;gBACL,OAAO;gBACP,YAAY;gBACZ,SAAS;gBACT,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,GAAG;gBACH,GAAG;gBACH,YAAY;aACb,CAAC;YACF,OAAO;YACP,SAAS;YACT,YAAY;YACZ,KAAK;YACL,WAAW;SACZ,CAAC,CAAC;QACH,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAClD,MAAM,oBAAoB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QAC1C,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,gBAAgB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACtC,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,MAAM,cAAc,EAAE,CAAC;IACzB,CAAC;AACH,CAAC"}
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
import type { DependencyRequirements } from '../eval-core/dependency-checker.js';
|
|
2
|
+
import type { Artifact, Sample, Task, VariantSpec } from '../types.js';
|
|
3
|
+
export interface PreparedEvaluationRun {
|
|
4
|
+
samples: Sample[];
|
|
5
|
+
artifacts: Artifact[];
|
|
6
|
+
tasks: Task[];
|
|
7
|
+
variantNames: string[];
|
|
8
|
+
requires?: DependencyRequirements;
|
|
9
|
+
}
|
|
10
|
+
export declare function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs, artifacts, dryRun, mcpConfig, }: {
|
|
11
|
+
samplesPath: string;
|
|
12
|
+
skillDir: string;
|
|
13
|
+
variantSpecs: VariantSpec[];
|
|
14
|
+
artifacts?: Artifact[];
|
|
15
|
+
dryRun: boolean;
|
|
16
|
+
mcpConfig?: string;
|
|
17
|
+
}): Promise<PreparedEvaluationRun>;
|
|
18
|
+
export declare function buildDryRunTaskReport({ model, judgeModel, executorName, samplesPath, skillDir, tasks, variantNames, }: {
|
|
19
|
+
model: string;
|
|
20
|
+
judgeModel: string;
|
|
21
|
+
executorName: string;
|
|
22
|
+
samplesPath: string;
|
|
23
|
+
skillDir: string;
|
|
24
|
+
tasks: Task[];
|
|
25
|
+
variantNames: string[];
|
|
26
|
+
}): {
|
|
27
|
+
dryRun: true;
|
|
28
|
+
model: string;
|
|
29
|
+
judgeModel: string;
|
|
30
|
+
variants: string[];
|
|
31
|
+
executor: string;
|
|
32
|
+
samplesPath: string;
|
|
33
|
+
skillDir: string;
|
|
34
|
+
totalTasks: number;
|
|
35
|
+
tasks: {
|
|
36
|
+
sample_id: string;
|
|
37
|
+
variant: string;
|
|
38
|
+
artifactKind: import("../types.js").ArtifactKind;
|
|
39
|
+
artifactSource: "baseline" | "variant-name" | "file-path" | "git" | "inline" | "custom";
|
|
40
|
+
executionStrategy: import("../types.js").ExecutionStrategyKind;
|
|
41
|
+
experimentType: import("../types.js").ExperimentType;
|
|
42
|
+
experimentRole: import("../types.js").ExperimentRole;
|
|
43
|
+
cwd: string | null;
|
|
44
|
+
promptPreview: string;
|
|
45
|
+
hasRubric: boolean;
|
|
46
|
+
hasAssertions: boolean;
|
|
47
|
+
hasDimensions: boolean;
|
|
48
|
+
hasSystem: boolean;
|
|
49
|
+
}[];
|
|
50
|
+
};
|
|
51
|
+
export declare function buildDryRunEachArtifacts(skillEntries: Array<{
|
|
52
|
+
name: string;
|
|
53
|
+
skillPath: string;
|
|
54
|
+
samplesPath: string;
|
|
55
|
+
}>): {
|
|
56
|
+
artifacts: {
|
|
57
|
+
name: string;
|
|
58
|
+
samplesPath: string;
|
|
59
|
+
sampleCount: number;
|
|
60
|
+
taskCount: number;
|
|
61
|
+
}[];
|
|
62
|
+
totalTasks: number;
|
|
63
|
+
};
|
|
64
|
+
//# sourceMappingURL=evaluation-preparation.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-preparation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AACjF,OAAO,KAAK,EAAE,QAAQ,EAAc,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAEnF,MAAM,WAAW,qBAAqB;IACpC,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,CAAC,EAAE,sBAAsB,CAAC;CACnC;AAED,wBAAsB,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,GACV,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,WAAW,EAAE,CAAC;IAC5B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,EAAE,OAAO,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,OAAO,CAAC,qBAAqB,CAAC,CAyCjC;AAED,wBAAgB,qBAAqB,CAAC,EACpC,KAAK,EACL,UAAU,EACV,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GACb,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;;;;;;;;;;;;;;;;;;;;;;;;EA6BA;AAED,wBAAgB,wBAAwB,CAAC,YAAY,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAAC;;;;;;;;EAerH"}
|
|
@@ -0,0 +1,90 @@
|
|
|
1
|
+
import { resolve } from 'node:path';
|
|
2
|
+
import { buildTasksFromArtifacts } from '../eval-core/task-planner.js';
|
|
3
|
+
import { loadSamples } from '../inputs/load-samples.js';
|
|
4
|
+
import { resolveArtifacts } from '../inputs/skill-loader.js';
|
|
5
|
+
import { buildVariantConfig } from '../eval-core/execution-strategy.js';
|
|
6
|
+
import { loadMcpConfig, resolveMcpUrls } from '../inputs/mcp-resolver.js';
|
|
7
|
+
import { resolveUrls } from '../inputs/url-fetcher.js';
|
|
8
|
+
export async function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs, artifacts, dryRun, mcpConfig, }) {
|
|
9
|
+
const { samples, requires } = loadSamples(samplesPath);
|
|
10
|
+
// Build expressions from specs (preserving order) and resolve to artifacts.
|
|
11
|
+
const variantExpressions = variantSpecs.map((spec) => spec.expr);
|
|
12
|
+
const resolvedArtifacts = artifacts || resolveArtifacts(resolve(skillDir), variantExpressions);
|
|
13
|
+
// Attach experimentRole to each artifact by matching spec.name to artifact.name.
|
|
14
|
+
const roleByName = {};
|
|
15
|
+
for (const spec of variantSpecs)
|
|
16
|
+
roleByName[spec.name] = spec.role;
|
|
17
|
+
for (const artifact of resolvedArtifacts) {
|
|
18
|
+
if (artifact.experimentRole)
|
|
19
|
+
continue; // preserve if already set (e.g. each-workflow)
|
|
20
|
+
const role = roleByName[artifact.name];
|
|
21
|
+
if (role)
|
|
22
|
+
artifact.experimentRole = role;
|
|
23
|
+
}
|
|
24
|
+
if (!dryRun) {
|
|
25
|
+
const mcpServers = loadMcpConfig(mcpConfig);
|
|
26
|
+
const mcpResolved = mcpServers ? await resolveMcpUrls(samples, mcpServers) : new Set();
|
|
27
|
+
await resolveUrls(samples, mcpResolved);
|
|
28
|
+
}
|
|
29
|
+
if (resolvedArtifacts.length === 0) {
|
|
30
|
+
throw new Error(`未发现任何 variant。请检查:\n`
|
|
31
|
+
+ ` 1. skill 目录是否存在:${resolve(skillDir)}\n`
|
|
32
|
+
+ ` 2. 目录下是否有 .md 文件或含 SKILL.md 的子目录\n`
|
|
33
|
+
+ ` 3. 通过 --control / --treatment 显式声明 variant 与角色,或用 --config eval.yaml`);
|
|
34
|
+
}
|
|
35
|
+
const tasks = buildTasksFromArtifacts(samples, resolvedArtifacts);
|
|
36
|
+
const variantNames = resolvedArtifacts.map((artifact) => artifact.name);
|
|
37
|
+
return {
|
|
38
|
+
samples,
|
|
39
|
+
artifacts: resolvedArtifacts,
|
|
40
|
+
tasks,
|
|
41
|
+
variantNames,
|
|
42
|
+
requires,
|
|
43
|
+
};
|
|
44
|
+
}
|
|
45
|
+
export function buildDryRunTaskReport({ model, judgeModel, executorName, samplesPath, skillDir, tasks, variantNames, }) {
|
|
46
|
+
return {
|
|
47
|
+
dryRun: true,
|
|
48
|
+
model,
|
|
49
|
+
judgeModel,
|
|
50
|
+
variants: variantNames,
|
|
51
|
+
executor: executorName,
|
|
52
|
+
samplesPath,
|
|
53
|
+
skillDir,
|
|
54
|
+
totalTasks: tasks.length,
|
|
55
|
+
tasks: tasks.map((task) => {
|
|
56
|
+
const config = buildVariantConfig(task.artifact);
|
|
57
|
+
return {
|
|
58
|
+
sample_id: task.sample_id,
|
|
59
|
+
variant: task.variant,
|
|
60
|
+
artifactKind: task.artifact.kind,
|
|
61
|
+
artifactSource: task.artifact.source,
|
|
62
|
+
executionStrategy: config.executionStrategy,
|
|
63
|
+
experimentType: config.experimentType,
|
|
64
|
+
experimentRole: config.experimentRole,
|
|
65
|
+
cwd: task.cwd,
|
|
66
|
+
promptPreview: task.prompt.slice(0, 100),
|
|
67
|
+
hasRubric: Boolean(task.rubric),
|
|
68
|
+
hasAssertions: Boolean(task.assertions?.length),
|
|
69
|
+
hasDimensions: Boolean(task.dimensions && Object.keys(task.dimensions).length),
|
|
70
|
+
hasSystem: Boolean(task.artifactContent),
|
|
71
|
+
};
|
|
72
|
+
}),
|
|
73
|
+
};
|
|
74
|
+
}
|
|
75
|
+
export function buildDryRunEachArtifacts(skillEntries) {
|
|
76
|
+
const artifacts = skillEntries.map((entry) => {
|
|
77
|
+
const { samples } = loadSamples(entry.samplesPath);
|
|
78
|
+
return {
|
|
79
|
+
name: entry.name,
|
|
80
|
+
samplesPath: entry.samplesPath,
|
|
81
|
+
sampleCount: samples.length,
|
|
82
|
+
taskCount: samples.length * 2,
|
|
83
|
+
};
|
|
84
|
+
});
|
|
85
|
+
return {
|
|
86
|
+
artifacts,
|
|
87
|
+
totalTasks: artifacts.reduce((sum, artifact) => sum + artifact.taskCount, 0),
|
|
88
|
+
};
|
|
89
|
+
}
|
|
90
|
+
//# sourceMappingURL=evaluation-preparation.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-preparation.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,8BAA8B,CAAC;AACvE,OAAO,EAAE,WAAW,EAAE,MAAM,2BAA2B,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAC7D,OAAO,EAAE,kBAAkB,EAAE,MAAM,oCAAoC,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC1E,OAAO,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAYvD,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,GAQV;IACC,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IAEvD,4EAA4E;IAC5E,MAAM,kBAAkB,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjE,MAAM,iBAAiB,GAAG,SAAS,IAAI,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,kBAAkB,CAAC,CAAC;IAE/F,iFAAiF;IACjF,MAAM,UAAU,GAAwC,EAAE,CAAC;IAC3D,KAAK,MAAM,IAAI,IAAI,YAAY;QAAE,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC;IACnE,KAAK,MAAM,QAAQ,IAAI,iBAAiB,EAAE,CAAC;QACzC,IAAI,QAAQ,CAAC,cAAc;YAAE,SAAS,CAAE,+CAA+C;QACvF,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACvC,IAAI,IAAI;YAAE,QAAQ,CAAC,cAAc,GAAG,IAAI,CAAC;IAC3C,CAAC;IAED,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,UAAU,GAAsB,aAAa,CAAC,SAAS,CAAC,CAAC;QAC/D,MAAM,WAAW,GAAG,UAAU,CAAC,CAAC,CAAC,MAAM,cAAc,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAU,CAAC;QAC/F,MAAM,WAAW,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;IAC1C,CAAC;IAED,IAAI,iBAAiB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,IAAI,KAAK,CACb,sBAAsB;cACpB,qBAAqB,OAAO,CAAC,QAAQ,CAAC,IAAI;cAC1C,sCAAsC;cACtC,wEAAwE,CAC3E,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,uBAAuB,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;IAClE,MAAM,YAAY,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAExE,OAAO;QACL,OAAO;QACP,SAAS,EAAE,iBAAiB;QAC5B,KAAK;QACL,YAAY;QACZ,QAAQ;KACT,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,EACpC,KAAK,EACL,UAAU,EACV,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GASb;IACC,OAAO;QACL,MAAM,EAAE,IAAa;QACrB,KAAK;QACL,UAAU;QACV,QAAQ,EAAE,YAAY;QACtB,QAAQ,EAAE,YAAY;QACtB,WAAW;QACX,QAAQ;QACR,UAAU,EAAE,KAAK,CAAC,MAAM;QACxB,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACxB,MAAM,MAAM,GAAG,kBAAkB,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACjD,OAAO;gBACL,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,YAAY,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;gBAChC,cAAc,EAAE,IAAI,CAAC,QAAQ,CAAC,MAAM;gBACpC,iBAAiB,EAAE,MAAM,CAAC,iBAAiB;gBAC3C,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,GAAG,EAAE,IAAI,CAAC,GAAG;gBACb,aAAa,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;gBACxC,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC;gBAC/B,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,EAAE,MAAM,CAAC;gBAC/C,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC;gBAC9E,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,CAAC;aACzC,CAAC;QACJ,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,YAA6E;IACpH,MAAM,SAAS,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;QAC3C,MAAM,EAAE,OAAO,EAAE,GAAG,WAAW,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;QACnD,OAAO;YACL,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,OAAO,CAAC,MAAM,GAAG,CAAC;SAC9B,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,OAAO;QACL,SAAS;QACT,UAAU,EAAE,SAAS,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,QAAQ,EAAE,EAAE,CAAC,GAAG,GAAG,QAAQ,CAAC,SAAS,EAAE,CAAC,CAAC;KAC7E,CAAC;AACJ,CAAC"}
|
|
@@ -0,0 +1,111 @@
|
|
|
1
|
+
import type { Artifact, JobStore, ProgressCallback, Report, VarianceData, VariantSpec } from '../types.js';
|
|
2
|
+
export interface SkillProgressInfo {
|
|
3
|
+
phase: string;
|
|
4
|
+
skill: string;
|
|
5
|
+
current: number;
|
|
6
|
+
total: number;
|
|
7
|
+
}
|
|
8
|
+
interface CommonEvaluationOptions {
|
|
9
|
+
model?: string;
|
|
10
|
+
judgeModel?: string;
|
|
11
|
+
outputDir?: string | null;
|
|
12
|
+
project?: string;
|
|
13
|
+
owner?: string;
|
|
14
|
+
tags?: string[];
|
|
15
|
+
noJudge?: boolean;
|
|
16
|
+
concurrency?: number;
|
|
17
|
+
timeoutMs?: number;
|
|
18
|
+
executorName?: string;
|
|
19
|
+
judgeExecutorName?: string;
|
|
20
|
+
jobStore?: JobStore | null;
|
|
21
|
+
persistJob?: boolean;
|
|
22
|
+
onProgress?: ProgressCallback | null;
|
|
23
|
+
skipPreflight?: boolean;
|
|
24
|
+
mcpConfig?: string;
|
|
25
|
+
verbose?: boolean;
|
|
26
|
+
layeredStats?: boolean;
|
|
27
|
+
/** --repeat N. 1 表示单次(默认); > 1 时在 runMultiple 层聚合 variance。
|
|
28
|
+
* 记入 report.meta.request.repeat 让 meta 如实反映用户输入。 */
|
|
29
|
+
repeat?: number;
|
|
30
|
+
/** --each 模式标记, true 表示当前评测是 each 批量流程(每个 skill 独立对比 baseline)。
|
|
31
|
+
* 记入 report.meta.request.each。 */
|
|
32
|
+
each?: boolean;
|
|
33
|
+
}
|
|
34
|
+
export interface RunEvaluationOptions extends CommonEvaluationOptions {
|
|
35
|
+
samplesPath: string;
|
|
36
|
+
skillDir: string;
|
|
37
|
+
variantSpecs?: VariantSpec[];
|
|
38
|
+
artifacts?: Artifact[];
|
|
39
|
+
dryRun?: boolean;
|
|
40
|
+
blind?: boolean;
|
|
41
|
+
noCache?: boolean;
|
|
42
|
+
retry?: number;
|
|
43
|
+
resume?: string;
|
|
44
|
+
}
|
|
45
|
+
export interface RunEachEvaluationOptions extends CommonEvaluationOptions {
|
|
46
|
+
skillDir: string;
|
|
47
|
+
dryRun?: boolean;
|
|
48
|
+
onSkillProgress?: ((info: SkillProgressInfo) => void) | null;
|
|
49
|
+
}
|
|
50
|
+
export interface RunMultipleOptions extends RunEvaluationOptions {
|
|
51
|
+
repeat?: number;
|
|
52
|
+
onRepeatProgress?: ((info: {
|
|
53
|
+
run: number;
|
|
54
|
+
total: number;
|
|
55
|
+
}) => void) | null;
|
|
56
|
+
}
|
|
57
|
+
interface DryRunTask {
|
|
58
|
+
sample_id: string;
|
|
59
|
+
variant: string;
|
|
60
|
+
artifactKind: Artifact['kind'];
|
|
61
|
+
artifactSource: Artifact['source'];
|
|
62
|
+
executionStrategy: string;
|
|
63
|
+
experimentType: string;
|
|
64
|
+
experimentRole: string;
|
|
65
|
+
cwd: string | null;
|
|
66
|
+
promptPreview: string;
|
|
67
|
+
hasRubric: boolean;
|
|
68
|
+
hasAssertions: boolean;
|
|
69
|
+
hasDimensions: boolean;
|
|
70
|
+
hasSystem: boolean;
|
|
71
|
+
}
|
|
72
|
+
interface DryRunBase {
|
|
73
|
+
dryRun: true;
|
|
74
|
+
model: string;
|
|
75
|
+
judgeModel: string;
|
|
76
|
+
executor: string;
|
|
77
|
+
skillDir: string;
|
|
78
|
+
totalTasks: number;
|
|
79
|
+
}
|
|
80
|
+
export interface DryRunReport extends DryRunBase {
|
|
81
|
+
variants: string[];
|
|
82
|
+
samplesPath: string;
|
|
83
|
+
tasks: DryRunTask[];
|
|
84
|
+
}
|
|
85
|
+
export declare function runEvaluation({ samplesPath, skillDir, variantSpecs, artifacts, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, blind, concurrency, timeoutMs, noCache, executorName, judgeExecutorName, jobStore, persistJob, onProgress, skipPreflight, mcpConfig, verbose, retry, resume, layeredStats, repeat, each, }: RunEvaluationOptions): Promise<{
|
|
86
|
+
report: Report | DryRunReport;
|
|
87
|
+
filePath: string | null;
|
|
88
|
+
}>;
|
|
89
|
+
interface DryRunEachSkill {
|
|
90
|
+
name: string;
|
|
91
|
+
samplesPath: string;
|
|
92
|
+
sampleCount: number;
|
|
93
|
+
taskCount: number;
|
|
94
|
+
}
|
|
95
|
+
export interface DryRunEachReport extends DryRunBase {
|
|
96
|
+
each: true;
|
|
97
|
+
totalArtifacts: number;
|
|
98
|
+
artifacts: DryRunEachSkill[];
|
|
99
|
+
}
|
|
100
|
+
export declare function buildVarianceData(runs: Report[]): VarianceData | null;
|
|
101
|
+
export declare function runEachEvaluation({ skillDir, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, repeat, }: RunEachEvaluationOptions): Promise<{
|
|
102
|
+
report: Report | DryRunEachReport;
|
|
103
|
+
filePath: string | null;
|
|
104
|
+
}>;
|
|
105
|
+
export declare function runMultiple({ repeat, onRepeatProgress, ...config }: RunMultipleOptions): Promise<{
|
|
106
|
+
report: Report;
|
|
107
|
+
aggregated: VarianceData | null;
|
|
108
|
+
filePath: string | null;
|
|
109
|
+
}>;
|
|
110
|
+
export {};
|
|
111
|
+
//# sourceMappingURL=run-evaluation.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"run-evaluation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/run-evaluation.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EACV,QAAQ,EAER,QAAQ,EACR,gBAAgB,EAChB,MAAM,EAGN,YAAY,EAGZ,WAAW,EAGZ,MAAM,aAAa,CAAC;AAErB,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,UAAU,uBAAuB;IAC/B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAIlB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;yDACqD;IACrD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;uCACmC;IACnC,IAAI,CAAC,EAAE,OAAO,CAAC;CAChB;AAED,MAAM,WAAW,oBAAqB,SAAQ,uBAAuB;IACnE,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,WAAW,EAAE,CAAC;IAC7B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,wBAAyB,SAAQ,uBAAuB;IACvE,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE,iBAAiB,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;CAC9D;AAED,MAAM,WAAW,kBAAmB,SAAQ,oBAAoB;IAC9D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,gBAAgB,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;CAC5E;AAED,UAAU,UAAU;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC/B,cAAc,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACnC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,OAAO,CAAC;IACnB,aAAa,EAAE,OAAO,CAAC;IACvB,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,EAAE,OAAO,CAAC;CACpB;AAED,UAAU,UAAU;IAClB,MAAM,EAAE,IAAI,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,YAAa,SAAQ,UAAU;IAC9C,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,UAAU,EAAE,CAAC;CACrB;AAED,wBAAsB,aAAa,CAAC,EAClC,WAAW,EACX,QAAQ,EACR,YAAiB,EACjB,SAAS,EACT,KAAqB,EACrB,UAAwB,EACxB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,YAAuB,EACvB,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,KAAS,EACT,MAAM,EACN,YAAoB,EACpB,MAAM,EACN,IAAI,GACL,EAAE,oBAAoB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,GAAG,YAAY,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAgF5F;AAED,UAAU,eAAe;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,gBAAiB,SAAQ,UAAU;IAClD,IAAI,EAAE,IAAI,CAAC;IACX,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,eAAe,EAAE,CAAC;CAC9B;AA6CD,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,YAAY,GAAG,IAAI,CAuErE;AAED,wBAAsB,iBAAiB,CAAC,EACtC,QAAQ,EACR,KAAqB,EACrB,UAAwB,EACxB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,WAAe,EACf,SAAS,EACT,YAAuB,EACvB,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,MAAM,GACP,EAAE,wBAAwB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,GAAG,gBAAgB,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAuDpG;AAED,wBAAsB,WAAW,CAAC,EAAE,MAAU,EAAE,gBAAgB,EAAE,GAAG,MAAM,EAAE,EAAE,kBAAkB;;;;GAyBhG"}
|