oh-my-knowledge 0.0.1 → 0.18.0
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 +537 -332
- 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 +1004 -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 +48 -0
- package/dist/src/eval-core/evaluation-job.d.ts.map +1 -0
- package/dist/src/eval-core/evaluation-job.js +112 -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 +133 -0
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +1 -0
- package/dist/src/eval-workflows/each-evaluation-workflow.js +169 -0
- package/dist/src/eval-workflows/each-evaluation-workflow.js.map +1 -0
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts +38 -0
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -0
- package/dist/src/eval-workflows/evaluation-pipeline.js +214 -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 +105 -0
- package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -0
- package/dist/src/eval-workflows/run-evaluation.js +264 -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/production-analyzer.d.ts +61 -0
- package/dist/src/observability/production-analyzer.d.ts.map +1 -0
- package/dist/src/observability/production-analyzer.js +183 -0
- package/dist/src/observability/production-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 +341 -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 +338 -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 +476 -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 +218 -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 +16 -0
- package/dist/src/server/report-server.d.ts.map +1 -0
- package/dist/src/server/report-server.js +196 -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 +553 -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 +44 -3
|
@@ -0,0 +1,196 @@
|
|
|
1
|
+
import { createCache, cacheKey } from './cache.js';
|
|
2
|
+
import { buildVariantResult } from './schema.js';
|
|
3
|
+
import { grade } from '../grading/index.js';
|
|
4
|
+
import { checkFacts } from './fact-checker.js';
|
|
5
|
+
import { resolveExecutionStrategy } from './execution-strategy.js';
|
|
6
|
+
import { resolve, join, dirname } from 'node:path';
|
|
7
|
+
import { homedir } from 'node:os';
|
|
8
|
+
async function runWithConcurrency(tasks, concurrency, fn) {
|
|
9
|
+
let index = 0;
|
|
10
|
+
async function worker() {
|
|
11
|
+
while (index < tasks.length) {
|
|
12
|
+
const i = index++;
|
|
13
|
+
await fn(tasks[i]);
|
|
14
|
+
}
|
|
15
|
+
}
|
|
16
|
+
await Promise.all(Array.from({ length: Math.min(concurrency, tasks.length) }, () => worker()));
|
|
17
|
+
}
|
|
18
|
+
function makeErrorResult(error) {
|
|
19
|
+
const message = error instanceof Error ? error.message : String(error);
|
|
20
|
+
return {
|
|
21
|
+
ok: false,
|
|
22
|
+
output: null,
|
|
23
|
+
durationMs: 0,
|
|
24
|
+
durationApiMs: 0,
|
|
25
|
+
inputTokens: 0,
|
|
26
|
+
outputTokens: 0,
|
|
27
|
+
cacheReadTokens: 0,
|
|
28
|
+
cacheCreationTokens: 0,
|
|
29
|
+
costUSD: 0,
|
|
30
|
+
stopReason: 'error',
|
|
31
|
+
numTurns: 0,
|
|
32
|
+
error: message,
|
|
33
|
+
};
|
|
34
|
+
}
|
|
35
|
+
function sleep(ms) {
|
|
36
|
+
return new Promise((resolve) => setTimeout(resolve, ms));
|
|
37
|
+
}
|
|
38
|
+
export async function executeTasks({ tasks, executor, judgeExecutor, model, judgeModel, noJudge, samplesPath, concurrency, timeoutMs, noCache, verbose, onProgress, retry = 0, existingResults, }) {
|
|
39
|
+
const results = {};
|
|
40
|
+
let started = 0;
|
|
41
|
+
let completed = 0;
|
|
42
|
+
let skipped = 0;
|
|
43
|
+
let totalCostUSD = 0;
|
|
44
|
+
// Seed results from previous run (--resume)
|
|
45
|
+
if (existingResults) {
|
|
46
|
+
for (const [sampleId, variants] of Object.entries(existingResults)) {
|
|
47
|
+
results[sampleId] = { ...variants };
|
|
48
|
+
}
|
|
49
|
+
}
|
|
50
|
+
const cacheDir = join(homedir(), '.oh-my-knowledge', 'cache');
|
|
51
|
+
const cache = noCache ? null : createCache(cacheDir);
|
|
52
|
+
async function executeTask(task) {
|
|
53
|
+
// Skip if already have a successful result (--resume)
|
|
54
|
+
if (existingResults?.[task.sample_id]?.[task.variant]?.ok) {
|
|
55
|
+
skipped++;
|
|
56
|
+
started++;
|
|
57
|
+
completed++;
|
|
58
|
+
onProgress?.({ phase: 'done', completed, total: tasks.length, sample_id: task.sample_id, variant: task.variant, skipped: true });
|
|
59
|
+
return;
|
|
60
|
+
}
|
|
61
|
+
started++;
|
|
62
|
+
const idx = started;
|
|
63
|
+
const total = tasks.length;
|
|
64
|
+
onProgress?.({ phase: 'start', completed: idx, total, sample_id: task.sample_id, variant: task.variant });
|
|
65
|
+
const executionPlan = resolveExecutionStrategy(task, model, timeoutMs, verbose);
|
|
66
|
+
let execResult;
|
|
67
|
+
const key = cacheKey(model, executionPlan.cacheSystem, executionPlan.input.prompt, executionPlan.input.cwd);
|
|
68
|
+
const cached = cache?.get(key);
|
|
69
|
+
const execStart = Date.now();
|
|
70
|
+
if (cached) {
|
|
71
|
+
execResult = { ...cached, cached: true };
|
|
72
|
+
}
|
|
73
|
+
else {
|
|
74
|
+
// Execute with retry on failure
|
|
75
|
+
const maxAttempts = 1 + Math.max(0, retry);
|
|
76
|
+
for (let attempt = 1; attempt <= maxAttempts; attempt++) {
|
|
77
|
+
try {
|
|
78
|
+
execResult = await executor(executionPlan.input);
|
|
79
|
+
}
|
|
80
|
+
catch (err) {
|
|
81
|
+
execResult = makeErrorResult(err);
|
|
82
|
+
}
|
|
83
|
+
if (execResult.ok || attempt === maxAttempts)
|
|
84
|
+
break;
|
|
85
|
+
// Exponential backoff before retry
|
|
86
|
+
const backoffMs = Math.min(2 ** (attempt - 1) * 1000, 30000);
|
|
87
|
+
onProgress?.({ phase: 'retry', completed: idx, total, sample_id: task.sample_id, variant: task.variant, attempt, maxAttempts });
|
|
88
|
+
await sleep(backoffMs);
|
|
89
|
+
}
|
|
90
|
+
if (cache && execResult.ok)
|
|
91
|
+
cache.set(key, execResult);
|
|
92
|
+
}
|
|
93
|
+
const execMs = Date.now() - execStart;
|
|
94
|
+
totalCostUSD += execResult.costUSD;
|
|
95
|
+
if (verbose && onProgress) {
|
|
96
|
+
onProgress({
|
|
97
|
+
phase: 'exec_done',
|
|
98
|
+
strategy: executionPlan.strategy,
|
|
99
|
+
completed: idx,
|
|
100
|
+
total,
|
|
101
|
+
sample_id: task.sample_id,
|
|
102
|
+
variant: task.variant,
|
|
103
|
+
durationMs: execResult.durationMs,
|
|
104
|
+
inputTokens: execResult.inputTokens,
|
|
105
|
+
outputTokens: execResult.outputTokens,
|
|
106
|
+
costUSD: execResult.costUSD,
|
|
107
|
+
outputPreview: execResult.output ? execResult.output.slice(0, 200) : null,
|
|
108
|
+
});
|
|
109
|
+
}
|
|
110
|
+
let gradeResult = null;
|
|
111
|
+
let gradeMs = 0;
|
|
112
|
+
if (execResult.ok) {
|
|
113
|
+
const hasGradingCriteria = task.rubric || task.assertions?.length || (task.dimensions && Object.keys(task.dimensions).length);
|
|
114
|
+
if (hasGradingCriteria) {
|
|
115
|
+
if (verbose && onProgress) {
|
|
116
|
+
onProgress({
|
|
117
|
+
phase: 'grading',
|
|
118
|
+
strategy: executionPlan.strategy,
|
|
119
|
+
completed: idx,
|
|
120
|
+
total,
|
|
121
|
+
sample_id: task.sample_id,
|
|
122
|
+
variant: task.variant,
|
|
123
|
+
});
|
|
124
|
+
}
|
|
125
|
+
const gradeStart = Date.now();
|
|
126
|
+
try {
|
|
127
|
+
gradeResult = await grade({
|
|
128
|
+
output: execResult.output,
|
|
129
|
+
sample: task._sample,
|
|
130
|
+
executor: judgeExecutor,
|
|
131
|
+
judgeModel,
|
|
132
|
+
allowLlmJudge: !noJudge,
|
|
133
|
+
execMetrics: {
|
|
134
|
+
costUSD: execResult.costUSD,
|
|
135
|
+
durationMs: execResult.durationMs,
|
|
136
|
+
numTurns: execResult.numTurns,
|
|
137
|
+
toolCalls: execResult.toolCalls,
|
|
138
|
+
turns: execResult.turns,
|
|
139
|
+
},
|
|
140
|
+
samplesDir: dirname(resolve(samplesPath)),
|
|
141
|
+
});
|
|
142
|
+
}
|
|
143
|
+
catch (err) {
|
|
144
|
+
gradeResult = { compositeScore: 0 };
|
|
145
|
+
const msg = err instanceof Error ? err.message : String(err);
|
|
146
|
+
onProgress?.({ phase: 'error', completed: idx, total, sample_id: task.sample_id, variant: task.variant, error: `评分失败: ${msg}` });
|
|
147
|
+
}
|
|
148
|
+
gradeMs = Date.now() - gradeStart;
|
|
149
|
+
if (gradeResult.judgeCostUSD)
|
|
150
|
+
totalCostUSD += gradeResult.judgeCostUSD;
|
|
151
|
+
}
|
|
152
|
+
}
|
|
153
|
+
completed++;
|
|
154
|
+
onProgress?.({
|
|
155
|
+
phase: 'done',
|
|
156
|
+
strategy: executionPlan.strategy,
|
|
157
|
+
completed,
|
|
158
|
+
total,
|
|
159
|
+
sample_id: task.sample_id,
|
|
160
|
+
variant: task.variant,
|
|
161
|
+
durationMs: execResult.durationMs,
|
|
162
|
+
inputTokens: execResult.inputTokens,
|
|
163
|
+
outputTokens: execResult.outputTokens,
|
|
164
|
+
costUSD: execResult.costUSD,
|
|
165
|
+
score: gradeResult?.compositeScore,
|
|
166
|
+
});
|
|
167
|
+
let factCheck;
|
|
168
|
+
if (execResult.ok && execResult.output && task.cwd) {
|
|
169
|
+
factCheck = checkFacts(execResult.output, resolve(task.cwd));
|
|
170
|
+
}
|
|
171
|
+
if (!results[task.sample_id])
|
|
172
|
+
results[task.sample_id] = {};
|
|
173
|
+
results[task.sample_id][task.variant] = buildVariantResult(execResult, gradeResult, { execMs, gradeMs, factCheck });
|
|
174
|
+
}
|
|
175
|
+
try {
|
|
176
|
+
await runWithConcurrency(tasks, concurrency, executeTask);
|
|
177
|
+
}
|
|
178
|
+
finally {
|
|
179
|
+
if (cache)
|
|
180
|
+
cache.save();
|
|
181
|
+
}
|
|
182
|
+
return { results, totalCostUSD, skipped };
|
|
183
|
+
}
|
|
184
|
+
export async function preflight(executor, model, timeoutMs = 15000) {
|
|
185
|
+
const result = await executor({
|
|
186
|
+
model,
|
|
187
|
+
system: '',
|
|
188
|
+
prompt: 'hi',
|
|
189
|
+
cwd: process.cwd(),
|
|
190
|
+
timeoutMs,
|
|
191
|
+
});
|
|
192
|
+
if (!result.ok) {
|
|
193
|
+
throw new Error(`预检失败 [${model}]: ${result.error}`);
|
|
194
|
+
}
|
|
195
|
+
}
|
|
196
|
+
//# sourceMappingURL=evaluation-execution.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-execution.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-execution.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AACnD,OAAO,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AACjD,OAAO,EAAE,KAAK,EAAE,MAAM,qBAAqB,CAAC;AAC5C,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C,OAAO,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AA8BlC,KAAK,UAAU,kBAAkB,CAAI,KAAU,EAAE,WAAmB,EAAE,EAA8B;IAClG,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,UAAU,MAAM;QACnB,OAAO,KAAK,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAC5B,MAAM,CAAC,GAAG,KAAK,EAAE,CAAC;YAClB,MAAM,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;QACrB,CAAC;IACH,CAAC;IACD,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;AACjG,CAAC;AAED,SAAS,eAAe,CAAC,KAAc;IACrC,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACvE,OAAO;QACL,EAAE,EAAE,KAAK;QACT,MAAM,EAAE,IAAI;QACZ,UAAU,EAAE,CAAC;QACb,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,CAAC;QACd,YAAY,EAAE,CAAC;QACf,eAAe,EAAE,CAAC;QAClB,mBAAmB,EAAE,CAAC;QACtB,OAAO,EAAE,CAAC;QACV,UAAU,EAAE,OAAO;QACnB,QAAQ,EAAE,CAAC;QACX,KAAK,EAAE,OAAO;KACf,CAAC;AACJ,CAAC;AAED,SAAS,KAAK,CAAC,EAAU;IACvB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;AAC3D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,EACjC,KAAK,EACL,QAAQ,EACR,aAAa,EACb,KAAK,EACL,UAAU,EACV,OAAO,EACP,WAAW,EACX,WAAW,EACX,SAAS,EACT,OAAO,EACP,OAAO,EACP,UAAU,EACV,KAAK,GAAG,CAAC,EACT,eAAe,GACK;IACpB,MAAM,OAAO,GAAkD,EAAE,CAAC;IAClE,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,4CAA4C;IAC5C,IAAI,eAAe,EAAE,CAAC;QACpB,KAAK,MAAM,CAAC,QAAQ,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC,EAAE,CAAC;YACnE,OAAO,CAAC,QAAQ,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,CAAC;QACtC,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,OAAO,CAAC,CAAC;IAC9D,MAAM,KAAK,GAAyB,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC;IAE3E,KAAK,UAAU,WAAW,CAAC,IAAU;QACnC,sDAAsD;QACtD,IAAI,eAAe,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,EAAE,CAAC;YAC1D,OAAO,EAAE,CAAC;YACV,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,CAAC;YACZ,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;YACjI,OAAO;QACT,CAAC;QAED,OAAO,EAAE,CAAC;QACV,MAAM,GAAG,GAAG,OAAO,CAAC;QACpB,MAAM,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC;QAC3B,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QAE1G,MAAM,aAAa,GAAG,wBAAwB,CAAC,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,CAAC,CAAC;QAEhF,IAAI,UAAsB,CAAC;QAC3B,MAAM,GAAG,GAAG,QAAQ,CAAC,KAAK,EAAE,aAAa,CAAC,WAAW,EAAE,aAAa,CAAC,KAAK,CAAC,MAAM,EAAE,aAAa,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC5G,MAAM,MAAM,GAAG,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAC/B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,MAAM,EAAE,CAAC;YACX,UAAU,GAAG,EAAE,GAAG,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC;QAC3C,CAAC;aAAM,CAAC;YACN,gCAAgC;YAChC,MAAM,WAAW,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;YAC3C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;gBACxD,IAAI,CAAC;oBACH,UAAU,GAAG,MAAM,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC;gBACnD,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,UAAU,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;gBACpC,CAAC;gBACD,IAAI,UAAW,CAAC,EAAE,IAAI,OAAO,KAAK,WAAW;oBAAE,MAAM;gBACrD,mCAAmC;gBACnC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,GAAG,IAAI,EAAE,KAAK,CAAC,CAAC;gBAC7D,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC;gBAChI,MAAM,KAAK,CAAC,SAAS,CAAC,CAAC;YACzB,CAAC;YACD,IAAI,KAAK,IAAI,UAAW,CAAC,EAAE;gBAAE,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,UAAW,CAAC,CAAC;QAC3D,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QACtC,YAAY,IAAI,UAAW,CAAC,OAAO,CAAC;QAEpC,IAAI,OAAO,IAAI,UAAU,EAAE,CAAC;YAC1B,UAAU,CAAC;gBACT,KAAK,EAAE,WAAW;gBAClB,QAAQ,EAAE,aAAa,CAAC,QAAQ;gBAChC,SAAS,EAAE,GAAG;gBACd,KAAK;gBACL,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,UAAU,EAAE,UAAW,CAAC,UAAU;gBAClC,WAAW,EAAE,UAAW,CAAC,WAAW;gBACpC,YAAY,EAAE,UAAW,CAAC,YAAY;gBACtC,OAAO,EAAE,UAAW,CAAC,OAAO;gBAC5B,aAAa,EAAE,UAAW,CAAC,MAAM,CAAC,CAAC,CAAC,UAAW,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;aAC5E,CAAC,CAAC;QACL,CAAC;QAED,IAAI,WAAW,GAAuB,IAAI,CAAC;QAC3C,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,IAAI,UAAW,CAAC,EAAE,EAAE,CAAC;YACnB,MAAM,kBAAkB,GAAG,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,UAAU,EAAE,MAAM,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;YAC9H,IAAI,kBAAkB,EAAE,CAAC;gBACvB,IAAI,OAAO,IAAI,UAAU,EAAE,CAAC;oBAC1B,UAAU,CAAC;wBACT,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,aAAa,CAAC,QAAQ;wBAChC,SAAS,EAAE,GAAG;wBACd,KAAK;wBACL,SAAS,EAAE,IAAI,CAAC,SAAS;wBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;qBACtB,CAAC,CAAC;gBACL,CAAC;gBACD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;gBAC9B,IAAI,CAAC;oBACH,WAAW,GAAG,MAAM,KAAK,CAAC;wBACxB,MAAM,EAAE,UAAW,CAAC,MAAO;wBAC3B,MAAM,EAAE,IAAI,CAAC,OAAO;wBACpB,QAAQ,EAAE,aAAa;wBACvB,UAAU;wBACV,aAAa,EAAE,CAAC,OAAO;wBACvB,WAAW,EAAE;4BACX,OAAO,EAAE,UAAW,CAAC,OAAO;4BAC5B,UAAU,EAAE,UAAW,CAAC,UAAU;4BAClC,QAAQ,EAAE,UAAW,CAAC,QAAQ;4BAC9B,SAAS,EAAE,UAAW,CAAC,SAAS;4BAChC,KAAK,EAAE,UAAW,CAAC,KAAK;yBACzB;wBACD,UAAU,EAAE,OAAO,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;qBAC1C,CAAC,CAAC;gBACL,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,WAAW,GAAG,EAAE,cAAc,EAAE,CAAC,EAAE,CAAC;oBACpC,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;oBAC7D,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,SAAS,GAAG,EAAE,EAAE,CAAC,CAAC;gBACnI,CAAC;gBACD,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,UAAU,CAAC;gBAClC,IAAI,WAAW,CAAC,YAAY;oBAAE,YAAY,IAAI,WAAW,CAAC,YAAY,CAAC;YACzE,CAAC;QACH,CAAC;QAED,SAAS,EAAE,CAAC;QACZ,UAAU,EAAE,CAAC;YACX,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,aAAa,CAAC,QAAQ;YAChC,SAAS;YACT,KAAK;YACL,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,UAAU,EAAE,UAAW,CAAC,UAAU;YAClC,WAAW,EAAE,UAAW,CAAC,WAAW;YACpC,YAAY,EAAE,UAAW,CAAC,YAAY;YACtC,OAAO,EAAE,UAAW,CAAC,OAAO;YAC5B,KAAK,EAAE,WAAW,EAAE,cAAc;SACnC,CAAC,CAAC;QAEH,IAAI,SAAsC,CAAC;QAC3C,IAAI,UAAW,CAAC,EAAE,IAAI,UAAW,CAAC,MAAM,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC;YACrD,SAAS,GAAG,UAAU,CAAC,UAAW,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;QAChE,CAAC;QAED,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,EAAE,CAAC;QAC3D,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,kBAAkB,CAAC,UAAW,EAAE,WAAW,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;IACvH,CAAC;IAED,IAAI,CAAC;QACH,MAAM,kBAAkB,CAAC,KAAK,EAAE,WAAW,EAAE,WAAW,CAAC,CAAC;IAC5D,CAAC;YAAS,CAAC;QACT,IAAI,KAAK;YAAE,KAAK,CAAC,IAAI,EAAE,CAAC;IAC1B,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,CAAC;AAC5C,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,QAAoB,EAAE,KAAa,EAAE,YAAoB,KAAK;IAC5F,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC;QAC5B,KAAK;QACL,MAAM,EAAE,EAAE;QACV,MAAM,EAAE,IAAI;QACZ,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE;QAClB,SAAS;KACV,CAAC,CAAC;IACH,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,SAAS,KAAK,MAAM,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IACtD,CAAC;AACH,CAAC"}
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
import type { Artifact, EvaluationErrorCategory, EvaluationJob, EvaluationRequest, EvaluationRun } from '../types.js';
|
|
2
|
+
export declare function buildEvaluationRequest({ samplesPath, skillDir, artifacts, model, judgeModel, executor, judgeExecutor, noJudge, concurrency, timeoutMs, noCache, dryRun, blind, project, owner, tags, }: {
|
|
3
|
+
samplesPath: string;
|
|
4
|
+
skillDir: string;
|
|
5
|
+
artifacts: Artifact[];
|
|
6
|
+
model: string;
|
|
7
|
+
judgeModel: string | null;
|
|
8
|
+
executor: string;
|
|
9
|
+
judgeExecutor?: string | null;
|
|
10
|
+
noJudge: boolean;
|
|
11
|
+
concurrency: number;
|
|
12
|
+
timeoutMs?: number;
|
|
13
|
+
noCache: boolean;
|
|
14
|
+
dryRun: boolean;
|
|
15
|
+
blind: boolean;
|
|
16
|
+
project?: string;
|
|
17
|
+
owner?: string;
|
|
18
|
+
tags?: string[];
|
|
19
|
+
}): EvaluationRequest;
|
|
20
|
+
export declare function createEvaluationRun(runId: string, startedAt?: string): {
|
|
21
|
+
run: EvaluationRun;
|
|
22
|
+
startedAt: string;
|
|
23
|
+
};
|
|
24
|
+
export declare function finalizeEvaluationRun(run: EvaluationRun, finishedAt?: string): EvaluationRun;
|
|
25
|
+
export declare function failEvaluationRun(run: EvaluationRun, finishedAt?: string): EvaluationRun;
|
|
26
|
+
export declare function createQueuedJob({ jobId, request, createdAt, }: {
|
|
27
|
+
jobId: string;
|
|
28
|
+
request: EvaluationRequest;
|
|
29
|
+
createdAt?: string;
|
|
30
|
+
}): EvaluationJob;
|
|
31
|
+
export declare function markJobRunning(job: EvaluationJob, runId: string, startedAt?: string): EvaluationJob;
|
|
32
|
+
export declare function createSucceededJob({ jobId, runId, reportId, request, createdAt, startedAt, finishedAt, }: {
|
|
33
|
+
jobId: string;
|
|
34
|
+
runId: string;
|
|
35
|
+
reportId: string;
|
|
36
|
+
request: EvaluationRequest;
|
|
37
|
+
createdAt: string;
|
|
38
|
+
startedAt: string;
|
|
39
|
+
finishedAt: string;
|
|
40
|
+
}): EvaluationJob;
|
|
41
|
+
export declare function classifyEvaluationError(error: string): EvaluationErrorCategory;
|
|
42
|
+
export declare function createFailedJob({ job, error, errorCategory, finishedAt, }: {
|
|
43
|
+
job: EvaluationJob;
|
|
44
|
+
error: string;
|
|
45
|
+
errorCategory?: EvaluationErrorCategory;
|
|
46
|
+
finishedAt?: string;
|
|
47
|
+
}): EvaluationJob;
|
|
48
|
+
//# sourceMappingURL=evaluation-job.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-job.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-job.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,uBAAuB,EAAE,aAAa,EAAE,iBAAiB,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAMtH,wBAAgB,sBAAsB,CAAC,EACrC,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,QAAQ,EACR,aAAa,EACb,OAAO,EACP,WAAW,EACX,SAAS,EACT,OAAO,EACP,MAAM,EACN,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,GACL,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB,GAAG,iBAAiB,CAmBpB;AAED,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG;IAAE,GAAG,EAAE,aAAa,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAS1H;AAED,wBAAgB,qBAAqB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMtG;AAED,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMlG;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,OAAO,EACP,SAAoB,GACrB,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAQhB;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,aAAa,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG,aAAa,CAQ7G;AAED,wBAAgB,kBAAkB,CAAC,EACjC,KAAK,EACL,KAAK,EACL,QAAQ,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,UAAU,GACX,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAYhB;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,MAAM,GAAG,uBAAuB,CA4B9E;AAED,wBAAgB,eAAe,CAAC,EAC9B,GAAG,EACH,KAAK,EACL,aAA8C,EAC9C,UAAqB,GACtB,EAAE;IACD,GAAG,EAAE,aAAa,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,uBAAuB,CAAC;IACxC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,aAAa,CAShB"}
|
|
@@ -0,0 +1,112 @@
|
|
|
1
|
+
function nowIso() {
|
|
2
|
+
return new Date().toISOString();
|
|
3
|
+
}
|
|
4
|
+
export function buildEvaluationRequest({ samplesPath, skillDir, artifacts, model, judgeModel, executor, judgeExecutor, noJudge, concurrency, timeoutMs, noCache, dryRun, blind, project, owner, tags, }) {
|
|
5
|
+
return {
|
|
6
|
+
samplesPath,
|
|
7
|
+
skillDir,
|
|
8
|
+
artifacts,
|
|
9
|
+
model,
|
|
10
|
+
judgeModel,
|
|
11
|
+
executor,
|
|
12
|
+
judgeExecutor,
|
|
13
|
+
noJudge,
|
|
14
|
+
concurrency,
|
|
15
|
+
timeoutMs,
|
|
16
|
+
noCache,
|
|
17
|
+
dryRun,
|
|
18
|
+
blind,
|
|
19
|
+
project,
|
|
20
|
+
owner,
|
|
21
|
+
tags,
|
|
22
|
+
};
|
|
23
|
+
}
|
|
24
|
+
export function createEvaluationRun(runId, startedAt = nowIso()) {
|
|
25
|
+
return {
|
|
26
|
+
startedAt,
|
|
27
|
+
run: {
|
|
28
|
+
runId,
|
|
29
|
+
startedAt,
|
|
30
|
+
status: 'running',
|
|
31
|
+
},
|
|
32
|
+
};
|
|
33
|
+
}
|
|
34
|
+
export function finalizeEvaluationRun(run, finishedAt = nowIso()) {
|
|
35
|
+
return {
|
|
36
|
+
...run,
|
|
37
|
+
finishedAt,
|
|
38
|
+
status: 'succeeded',
|
|
39
|
+
};
|
|
40
|
+
}
|
|
41
|
+
export function failEvaluationRun(run, finishedAt = nowIso()) {
|
|
42
|
+
return {
|
|
43
|
+
...run,
|
|
44
|
+
finishedAt,
|
|
45
|
+
status: 'failed',
|
|
46
|
+
};
|
|
47
|
+
}
|
|
48
|
+
export function createQueuedJob({ jobId, request, createdAt = nowIso(), }) {
|
|
49
|
+
return {
|
|
50
|
+
jobId,
|
|
51
|
+
status: 'queued',
|
|
52
|
+
createdAt,
|
|
53
|
+
updatedAt: createdAt,
|
|
54
|
+
request,
|
|
55
|
+
};
|
|
56
|
+
}
|
|
57
|
+
export function markJobRunning(job, runId, startedAt = nowIso()) {
|
|
58
|
+
return {
|
|
59
|
+
...job,
|
|
60
|
+
status: 'running',
|
|
61
|
+
runId,
|
|
62
|
+
startedAt,
|
|
63
|
+
updatedAt: startedAt,
|
|
64
|
+
};
|
|
65
|
+
}
|
|
66
|
+
export function createSucceededJob({ jobId, runId, reportId, request, createdAt, startedAt, finishedAt, }) {
|
|
67
|
+
return {
|
|
68
|
+
jobId,
|
|
69
|
+
status: 'succeeded',
|
|
70
|
+
createdAt,
|
|
71
|
+
startedAt,
|
|
72
|
+
finishedAt,
|
|
73
|
+
updatedAt: finishedAt,
|
|
74
|
+
request,
|
|
75
|
+
runId,
|
|
76
|
+
resultReportId: reportId,
|
|
77
|
+
};
|
|
78
|
+
}
|
|
79
|
+
export function classifyEvaluationError(error) {
|
|
80
|
+
const normalized = error.toLowerCase();
|
|
81
|
+
if (normalized.includes('missing required field')
|
|
82
|
+
|| normalized.includes('invalid samples file')
|
|
83
|
+
|| normalized.includes('skill not found')
|
|
84
|
+
|| normalized.includes('samples file')
|
|
85
|
+
|| normalized.includes('未发现任何 skill')) {
|
|
86
|
+
return 'user';
|
|
87
|
+
}
|
|
88
|
+
if (normalized.includes('judge')
|
|
89
|
+
|| normalized.includes('semantic_similarity')
|
|
90
|
+
|| normalized.includes('failed to parse judge response')) {
|
|
91
|
+
return 'judge';
|
|
92
|
+
}
|
|
93
|
+
if (normalized.includes('timed out')
|
|
94
|
+
|| normalized.includes('api request timed out')
|
|
95
|
+
|| normalized.includes('execution timed out')
|
|
96
|
+
|| normalized.includes('mcp')
|
|
97
|
+
|| normalized.includes('network')) {
|
|
98
|
+
return 'executor';
|
|
99
|
+
}
|
|
100
|
+
return 'system';
|
|
101
|
+
}
|
|
102
|
+
export function createFailedJob({ job, error, errorCategory = classifyEvaluationError(error), finishedAt = nowIso(), }) {
|
|
103
|
+
return {
|
|
104
|
+
...job,
|
|
105
|
+
status: 'failed',
|
|
106
|
+
finishedAt,
|
|
107
|
+
updatedAt: finishedAt,
|
|
108
|
+
error,
|
|
109
|
+
errorCategory,
|
|
110
|
+
};
|
|
111
|
+
}
|
|
112
|
+
//# sourceMappingURL=evaluation-job.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-job.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-job.ts"],"names":[],"mappings":"AAEA,SAAS,MAAM;IACb,OAAO,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,EACrC,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,QAAQ,EACR,aAAa,EACb,OAAO,EACP,WAAW,EACX,SAAS,EACT,OAAO,EACP,MAAM,EACN,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,GAkBL;IACC,OAAO;QACL,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU;QACV,QAAQ;QACR,aAAa;QACb,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM;QACN,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;KACL,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,KAAa,EAAE,YAAoB,MAAM,EAAE;IAC7E,OAAO;QACL,SAAS;QACT,GAAG,EAAE;YACH,KAAK;YACL,SAAS;YACT,MAAM,EAAE,SAAS;SAClB;KACF,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,GAAkB,EAAE,aAAqB,MAAM,EAAE;IACrF,OAAO;QACL,GAAG,GAAG;QACN,UAAU;QACV,MAAM,EAAE,WAAW;KACpB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,GAAkB,EAAE,aAAqB,MAAM,EAAE;IACjF,OAAO;QACL,GAAG,GAAG;QACN,UAAU;QACV,MAAM,EAAE,QAAQ;KACjB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,KAAK,EACL,OAAO,EACP,SAAS,GAAG,MAAM,EAAE,GAKrB;IACC,OAAO;QACL,KAAK;QACL,MAAM,EAAE,QAAQ;QAChB,SAAS;QACT,SAAS,EAAE,SAAS;QACpB,OAAO;KACR,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,GAAkB,EAAE,KAAa,EAAE,YAAoB,MAAM,EAAE;IAC5F,OAAO;QACL,GAAG,GAAG;QACN,MAAM,EAAE,SAAS;QACjB,KAAK;QACL,SAAS;QACT,SAAS,EAAE,SAAS;KACrB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,EACjC,KAAK,EACL,KAAK,EACL,QAAQ,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,UAAU,GASX;IACC,OAAO;QACL,KAAK;QACL,MAAM,EAAE,WAAW;QACnB,SAAS;QACT,SAAS;QACT,UAAU;QACV,SAAS,EAAE,UAAU;QACrB,OAAO;QACP,KAAK;QACL,cAAc,EAAE,QAAQ;KACzB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAa;IACnD,MAAM,UAAU,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;IACvC,IACE,UAAU,CAAC,QAAQ,CAAC,wBAAwB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,sBAAsB,CAAC;WAC3C,UAAU,CAAC,QAAQ,CAAC,iBAAiB,CAAC;WACtC,UAAU,CAAC,QAAQ,CAAC,cAAc,CAAC;WACnC,UAAU,CAAC,QAAQ,CAAC,aAAa,CAAC,EACrC,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,IACE,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC;WACzB,UAAU,CAAC,QAAQ,CAAC,qBAAqB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,gCAAgC,CAAC,EACxD,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;IACD,IACE,UAAU,CAAC,QAAQ,CAAC,WAAW,CAAC;WAC7B,UAAU,CAAC,QAAQ,CAAC,uBAAuB,CAAC;WAC5C,UAAU,CAAC,QAAQ,CAAC,qBAAqB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,KAAK,CAAC;WAC1B,UAAU,CAAC,QAAQ,CAAC,SAAS,CAAC,EACjC,CAAC;QACD,OAAO,UAAU,CAAC;IACpB,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,GAAG,EACH,KAAK,EACL,aAAa,GAAG,uBAAuB,CAAC,KAAK,CAAC,EAC9C,UAAU,GAAG,MAAM,EAAE,GAMtB;IACC,OAAO;QACL,GAAG,GAAG;QACN,MAAM,EAAE,QAAQ;QAChB,UAAU;QACV,SAAS,EAAE,UAAU;QACrB,KAAK;QACL,aAAa;KACd,CAAC;AACJ,CAAC"}
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
import type { Artifact, Report, Sample, Task, VariantResult, EvaluationJob, EvaluationRequest, EvaluationRun } from '../types.js';
|
|
2
|
+
export declare const DEFAULT_OUTPUT_DIR: string;
|
|
3
|
+
interface AggregateReportOptions {
|
|
4
|
+
runId: string;
|
|
5
|
+
variants: string[];
|
|
6
|
+
model: string;
|
|
7
|
+
judgeModel: string;
|
|
8
|
+
noJudge: boolean;
|
|
9
|
+
executorName: string;
|
|
10
|
+
samples: Sample[];
|
|
11
|
+
tasks: Task[];
|
|
12
|
+
results: Record<string, Record<string, VariantResult>>;
|
|
13
|
+
totalCostUSD: number;
|
|
14
|
+
artifacts: Artifact[];
|
|
15
|
+
request?: EvaluationRequest;
|
|
16
|
+
run?: EvaluationRun;
|
|
17
|
+
job?: EvaluationJob;
|
|
18
|
+
layeredStats?: boolean;
|
|
19
|
+
}
|
|
20
|
+
export declare function aggregateReport({ runId, variants, model, judgeModel, noJudge, executorName, samples, tasks, results, totalCostUSD, artifacts, request, run, job, layeredStats, }: AggregateReportOptions): Report;
|
|
21
|
+
export declare function applyBlindMode(report: Report, variants: string[], blindSeed: string): void;
|
|
22
|
+
export interface PersistableReport {
|
|
23
|
+
id: string;
|
|
24
|
+
}
|
|
25
|
+
export declare function persistReport(report: PersistableReport, outputDir: string | null): string | null;
|
|
26
|
+
export declare function generateRunId(variants: string[]): string;
|
|
27
|
+
export {};
|
|
28
|
+
//# sourceMappingURL=evaluation-reporting.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-reporting.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EACV,QAAQ,EACR,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EAGb,aAAa,EACb,iBAAiB,EACjB,aAAa,EACd,MAAM,aAAa,CAAC;AAgBrB,eAAO,MAAM,kBAAkB,EAAE,MAAuD,CAAC;AAiBzF,UAAU,sBAAsB;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IACvD,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACb,EAAE,sBAAsB,GAAG,MAAM,CAsCjC;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,CAmC1F;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,wBAAgB,aAAa,CAAC,MAAM,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,GAAG,IAAI,CAMhG;AAED,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,MAAM,CASxD"}
|
|
@@ -0,0 +1,123 @@
|
|
|
1
|
+
import { readFileSync, writeFileSync, mkdirSync, existsSync } from 'node:fs';
|
|
2
|
+
import { join, dirname } from 'node:path';
|
|
3
|
+
import { execFileSync } from 'node:child_process';
|
|
4
|
+
import { homedir } from 'node:os';
|
|
5
|
+
import { createHash } from 'node:crypto';
|
|
6
|
+
import { fileURLToPath } from 'node:url';
|
|
7
|
+
import { buildVariantSummary } from './schema.js';
|
|
8
|
+
import { buildVariantConfig } from './execution-strategy.js';
|
|
9
|
+
const __dirname = dirname(fileURLToPath(import.meta.url));
|
|
10
|
+
function findPackageJson(startDir) {
|
|
11
|
+
let dir = startDir;
|
|
12
|
+
for (let i = 0; i < 5; i++) {
|
|
13
|
+
const candidate = join(dir, 'package.json');
|
|
14
|
+
if (existsSync(candidate))
|
|
15
|
+
return candidate;
|
|
16
|
+
dir = dirname(dir);
|
|
17
|
+
}
|
|
18
|
+
return join(startDir, '..', 'package.json');
|
|
19
|
+
}
|
|
20
|
+
const PKG = JSON.parse(readFileSync(findPackageJson(__dirname), 'utf-8'));
|
|
21
|
+
export const DEFAULT_OUTPUT_DIR = join(homedir(), '.oh-my-knowledge', 'reports');
|
|
22
|
+
function hashString(str) {
|
|
23
|
+
return createHash('sha256').update(str).digest('hex').slice(0, 12);
|
|
24
|
+
}
|
|
25
|
+
function getGitInfo() {
|
|
26
|
+
try {
|
|
27
|
+
const commit = execFileSync('git', ['rev-parse', 'HEAD'], { encoding: 'utf-8' }).trim();
|
|
28
|
+
const branch = execFileSync('git', ['rev-parse', '--abbrev-ref', 'HEAD'], { encoding: 'utf-8' }).trim();
|
|
29
|
+
const dirty = execFileSync('git', ['status', '--porcelain'], { encoding: 'utf-8' }).trim().length > 0;
|
|
30
|
+
return { commit, commitShort: commit.slice(0, 7), branch, dirty };
|
|
31
|
+
}
|
|
32
|
+
catch {
|
|
33
|
+
return null;
|
|
34
|
+
}
|
|
35
|
+
}
|
|
36
|
+
export function aggregateReport({ runId, variants, model, judgeModel, noJudge, executorName, samples, tasks, results, totalCostUSD, artifacts, request, run, job, layeredStats, }) {
|
|
37
|
+
const summary = {};
|
|
38
|
+
for (const variant of variants) {
|
|
39
|
+
const entries = Object.values(results).map((result) => result[variant]).filter(Boolean);
|
|
40
|
+
summary[variant] = buildVariantSummary(entries);
|
|
41
|
+
}
|
|
42
|
+
const artifactHashes = Object.fromEntries(artifacts.map((artifact) => [artifact.name, artifact.content ? hashString(artifact.content) : 'no-skill']));
|
|
43
|
+
return {
|
|
44
|
+
id: runId,
|
|
45
|
+
meta: {
|
|
46
|
+
variants,
|
|
47
|
+
model,
|
|
48
|
+
judgeModel: noJudge ? null : judgeModel,
|
|
49
|
+
executor: executorName,
|
|
50
|
+
sampleCount: samples.length,
|
|
51
|
+
taskCount: tasks.length,
|
|
52
|
+
totalCostUSD: Number(totalCostUSD.toFixed(6)),
|
|
53
|
+
timestamp: new Date().toISOString(),
|
|
54
|
+
cliVersion: PKG.version,
|
|
55
|
+
nodeVersion: process.version,
|
|
56
|
+
artifactHashes,
|
|
57
|
+
variantConfigs: artifacts.map((artifact) => buildVariantConfig(artifact)),
|
|
58
|
+
request,
|
|
59
|
+
run,
|
|
60
|
+
job,
|
|
61
|
+
gitInfo: getGitInfo(),
|
|
62
|
+
...(layeredStats ? { layeredStats: true } : {}),
|
|
63
|
+
},
|
|
64
|
+
summary,
|
|
65
|
+
results: Object.entries(results).map(([sample_id, variantData]) => ({
|
|
66
|
+
sample_id,
|
|
67
|
+
variants: variantData,
|
|
68
|
+
})),
|
|
69
|
+
};
|
|
70
|
+
}
|
|
71
|
+
export function applyBlindMode(report, variants, blindSeed) {
|
|
72
|
+
const labels = variants.map((_, i) => String.fromCharCode(65 + i));
|
|
73
|
+
let seed = parseInt(hashString(blindSeed).slice(0, 8), 16) | 0;
|
|
74
|
+
const seededRandom = () => {
|
|
75
|
+
seed |= 0;
|
|
76
|
+
seed = seed + 0x6D2B79F5 | 0;
|
|
77
|
+
let value = Math.imul(seed ^ seed >>> 15, 1 | seed);
|
|
78
|
+
value ^= value + Math.imul(value ^ value >>> 7, 61 | value);
|
|
79
|
+
return ((value ^ value >>> 14) >>> 0) / 4294967296;
|
|
80
|
+
};
|
|
81
|
+
const shuffled = [...variants];
|
|
82
|
+
for (let i = shuffled.length - 1; i > 0; i--) {
|
|
83
|
+
const j = Math.floor(seededRandom() * (i + 1));
|
|
84
|
+
[shuffled[i], shuffled[j]] = [shuffled[j], shuffled[i]];
|
|
85
|
+
}
|
|
86
|
+
const blindMap = Object.fromEntries(shuffled.map((variant, i) => [labels[i], variant]));
|
|
87
|
+
const reverseMap = Object.fromEntries(Object.entries(blindMap).map(([label, variant]) => [variant, label]));
|
|
88
|
+
report.meta.blind = true;
|
|
89
|
+
report.meta.blindMap = blindMap;
|
|
90
|
+
report.meta.variants = labels;
|
|
91
|
+
const newSummary = {};
|
|
92
|
+
for (const [variant, stats] of Object.entries(report.summary)) {
|
|
93
|
+
newSummary[reverseMap[variant]] = stats;
|
|
94
|
+
}
|
|
95
|
+
report.summary = newSummary;
|
|
96
|
+
for (const result of report.results) {
|
|
97
|
+
const newVariants = {};
|
|
98
|
+
for (const [variant, data] of Object.entries(result.variants)) {
|
|
99
|
+
newVariants[reverseMap[variant]] = data;
|
|
100
|
+
}
|
|
101
|
+
result.variants = newVariants;
|
|
102
|
+
}
|
|
103
|
+
}
|
|
104
|
+
export function persistReport(report, outputDir) {
|
|
105
|
+
if (!outputDir)
|
|
106
|
+
return null;
|
|
107
|
+
if (!existsSync(outputDir))
|
|
108
|
+
mkdirSync(outputDir, { recursive: true });
|
|
109
|
+
const filePath = join(outputDir, `${report.id}.json`);
|
|
110
|
+
writeFileSync(filePath, JSON.stringify(report, null, 2));
|
|
111
|
+
return filePath;
|
|
112
|
+
}
|
|
113
|
+
export function generateRunId(variants) {
|
|
114
|
+
const d = new Date();
|
|
115
|
+
const pad = (n) => String(n).padStart(2, '0');
|
|
116
|
+
const date = `${d.getFullYear()}${pad(d.getMonth() + 1)}${pad(d.getDate())}`;
|
|
117
|
+
const time = `${pad(d.getHours())}${pad(d.getMinutes())}`;
|
|
118
|
+
const variantPart = variants
|
|
119
|
+
.map((variant) => variant.replaceAll(/[\\/:]/g, '-').replaceAll(/[^a-zA-Z0-9._@-]/g, '_'))
|
|
120
|
+
.join('-vs-');
|
|
121
|
+
return `${variantPart}-${date}-${time}`;
|
|
122
|
+
}
|
|
123
|
+
//# sourceMappingURL=evaluation-reporting.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-reporting.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,SAAS,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAClD,OAAO,EAAE,kBAAkB,EAAE,MAAM,yBAAyB,CAAC;AAc7D,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE1D,SAAS,eAAe,CAAC,QAAgB;IACvC,IAAI,GAAG,GAAG,QAAQ,CAAC;IACnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC;QAC5C,IAAI,UAAU,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;QAC5C,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC,QAAQ,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC;AAC9C,CAAC;AAED,MAAM,GAAG,GAAwB,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,eAAe,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,CAAwB,CAAC;AAEtH,MAAM,CAAC,MAAM,kBAAkB,GAAW,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,SAAS,CAAC,CAAC;AAEzF,SAAS,UAAU,CAAC,GAAW;IAC7B,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACrE,CAAC;AAED,SAAS,UAAU;IACjB,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxF,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,cAAc,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxG,MAAM,KAAK,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,aAAa,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;QACtG,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;IACpE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAoBD,MAAM,UAAU,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACW;IACvB,MAAM,OAAO,GAAmC,EAAE,CAAC;IACnD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACxF,OAAO,CAAC,OAAO,CAAC,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;IAClD,CAAC;IAED,MAAM,cAAc,GAAG,MAAM,CAAC,WAAW,CACvC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAC3G,CAAC;IAEF,OAAO;QACL,EAAE,EAAE,KAAK;QACT,IAAI,EAAE;YACJ,QAAQ;YACR,KAAK;YACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;YACvC,QAAQ,EAAE,YAAY;YACtB,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,KAAK,CAAC,MAAM;YACvB,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAC7C,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,UAAU,EAAE,GAAG,CAAC,OAAO;YACvB,WAAW,EAAE,OAAO,CAAC,OAAO;YAC5B,cAAc;YACd,cAAc,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,kBAAkB,CAAC,QAAQ,CAAC,CAAC;YACzE,OAAO;YACP,GAAG;YACH,GAAG;YACH,OAAO,EAAE,UAAU,EAAE;YACrB,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChD;QACD,OAAO;QACP,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,EAAE,CAAC,CAAC;YAClE,SAAS;YACT,QAAQ,EAAE,WAAW;SACtB,CAAC,CAAC;KACJ,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,QAAkB,EAAE,SAAiB;IAClF,MAAM,MAAM,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,YAAY,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC;IACnE,IAAI,IAAI,GAAG,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC;IAC/D,MAAM,YAAY,GAAG,GAAW,EAAE;QAChC,IAAI,IAAI,CAAC,CAAC;QACV,IAAI,GAAG,IAAI,GAAG,UAAU,GAAG,CAAC,CAAC;QAC7B,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,KAAK,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QACpD,KAAK,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,KAAK,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,CAAC;QAC5D,OAAO,CAAC,CAAC,KAAK,GAAG,KAAK,KAAK,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IACrD,CAAC,CAAC;IACF,MAAM,QAAQ,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC;IAC/B,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC/C,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1D,CAAC;IACD,MAAM,QAAQ,GAA2B,MAAM,CAAC,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC;IAChH,MAAM,UAAU,GAA2B,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IAEpI,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;IAChC,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,MAAM,CAAC;IAE9B,MAAM,UAAU,GAAmC,EAAE,CAAC;IACtD,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;QAC9D,UAAU,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,KAAK,CAAC;IAC1C,CAAC;IACD,MAAM,CAAC,OAAO,GAAG,UAAU,CAAC;IAE5B,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpC,MAAM,WAAW,GAAkC,EAAE,CAAC;QACtD,KAAK,MAAM,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC9D,WAAW,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,IAAI,CAAC;QAC1C,CAAC;QACD,MAAM,CAAC,QAAQ,GAAG,WAAW,CAAC;IAChC,CAAC;AACH,CAAC;AAMD,MAAM,UAAU,aAAa,CAAC,MAAyB,EAAE,SAAwB;IAC/E,IAAI,CAAC,SAAS;QAAE,OAAO,IAAI,CAAC;IAC5B,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,GAAG,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC;IACtD,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACzD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAkB;IAC9C,MAAM,CAAC,GAAG,IAAI,IAAI,EAAE,CAAC;IACrB,MAAM,GAAG,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC9D,MAAM,IAAI,GAAG,GAAG,CAAC,CAAC,WAAW,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,CAAC;IAC7E,MAAM,IAAI,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,UAAU,EAAE,CAAC,EAAE,CAAC;IAC1D,MAAM,WAAW,GAAG,QAAQ;SACzB,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,UAAU,CAAC,mBAAmB,EAAE,GAAG,CAAC,CAAC;SACzF,IAAI,CAAC,MAAM,CAAC,CAAC;IAChB,OAAO,GAAG,WAAW,IAAI,IAAI,IAAI,IAAI,EAAE,CAAC;AAC1C,CAAC"}
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
import type { Artifact, ExecutionStrategyKind, ExecutorInput, ExperimentType, Task, VariantConfig } from '../types.js';
|
|
2
|
+
export interface ExecutionPlan {
|
|
3
|
+
strategy: ExecutionStrategyKind;
|
|
4
|
+
cacheSystem: string;
|
|
5
|
+
input: ExecutorInput;
|
|
6
|
+
}
|
|
7
|
+
export declare function resolveArtifactExecutionStrategy(artifact: Artifact): ExecutionStrategyKind;
|
|
8
|
+
export declare function resolveExperimentType(artifact: Artifact): ExperimentType;
|
|
9
|
+
export declare function buildVariantConfig(artifact: Artifact): VariantConfig;
|
|
10
|
+
export declare function resolveExecutionStrategy(task: Task, model: string, timeoutMs?: number, verbose?: boolean): ExecutionPlan;
|
|
11
|
+
//# sourceMappingURL=execution-strategy.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"execution-strategy.d.ts","sourceRoot":"","sources":["../../../src/eval-core/execution-strategy.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,QAAQ,EAAE,qBAAqB,EAAE,aAAa,EAAE,cAAc,EAAE,IAAI,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAEvH,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,qBAAqB,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,aAAa,CAAC;CACtB;AAOD,wBAAgB,gCAAgC,CAAC,QAAQ,EAAE,QAAQ,GAAG,qBAAqB,CAc1F;AAED,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,QAAQ,GAAG,cAAc,CAIxE;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,QAAQ,GAAG,aAAa,CAyBpE;AAOD,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,aAAa,CA+DxH"}
|