oh-my-knowledge 0.23.0 → 0.24.0
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/README.md +63 -64
- package/README.zh.md +64 -63
- package/dist/src/analysis/report-diagnostics.d.ts +3 -2
- package/dist/src/analysis/report-diagnostics.d.ts.map +1 -1
- package/dist/src/analysis/report-diagnostics.js +107 -80
- package/dist/src/analysis/report-diagnostics.js.map +1 -1
- package/dist/src/analysis/sample-diagnostics.d.ts +4 -1
- package/dist/src/analysis/sample-diagnostics.d.ts.map +1 -1
- package/dist/src/analysis/sample-diagnostics.js +123 -29
- package/dist/src/analysis/sample-diagnostics.js.map +1 -1
- package/dist/src/authoring/evolver.d.ts +5 -0
- package/dist/src/authoring/evolver.d.ts.map +1 -1
- package/dist/src/authoring/evolver.js +20 -3
- package/dist/src/authoring/evolver.js.map +1 -1
- package/dist/src/cli/i18n-dict.d.ts +1 -1
- package/dist/src/cli/i18n-dict.d.ts.map +1 -1
- package/dist/src/cli/i18n-dict.js +28 -24
- package/dist/src/cli/i18n-dict.js.map +1 -1
- package/dist/src/cli/index.js +80 -66
- package/dist/src/cli/index.js.map +1 -1
- package/dist/src/cli/parse-run-config.d.ts.map +1 -1
- package/dist/src/cli/parse-run-config.js +4 -4
- package/dist/src/cli/parse-run-config.js.map +1 -1
- package/dist/src/cli/progress.d.ts +1 -0
- package/dist/src/cli/progress.d.ts.map +1 -1
- package/dist/src/cli/progress.js +8 -1
- package/dist/src/cli/progress.js.map +1 -1
- package/dist/src/eval-core/cache.d.ts +7 -5
- package/dist/src/eval-core/cache.d.ts.map +1 -1
- package/dist/src/eval-core/cache.js +11 -7
- package/dist/src/eval-core/cache.js.map +1 -1
- package/dist/src/eval-core/comparability.d.ts +11 -0
- package/dist/src/eval-core/comparability.d.ts.map +1 -0
- package/dist/src/eval-core/comparability.js +271 -0
- package/dist/src/eval-core/comparability.js.map +1 -0
- package/dist/src/eval-core/evaluation-execution.d.ts +4 -1
- package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-execution.js +25 -17
- package/dist/src/eval-core/evaluation-execution.js.map +1 -1
- package/dist/src/eval-core/evaluation-job.d.ts +2 -2
- package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-job.js +2 -2
- package/dist/src/eval-core/evaluation-job.js.map +1 -1
- package/dist/src/eval-core/evaluation-reporting.d.ts +3 -1
- package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-reporting.js +64 -4
- package/dist/src/eval-core/evaluation-reporting.js.map +1 -1
- package/dist/src/eval-core/execution-strategy.js +2 -2
- package/dist/src/eval-core/execution-strategy.js.map +1 -1
- package/dist/src/eval-core/schema.d.ts.map +1 -1
- package/dist/src/eval-core/schema.js +13 -0
- package/dist/src/eval-core/schema.js.map +1 -1
- package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts +111 -0
- package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts.map +1 -0
- package/dist/src/eval-workflows/batch-evaluation-workflow.js +215 -0
- package/dist/src/eval-workflows/batch-evaluation-workflow.js.map +1 -0
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts +5 -3
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
- package/dist/src/eval-workflows/evaluation-pipeline.js +6 -5
- package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
- package/dist/src/eval-workflows/evaluation-preparation.d.ts +8 -6
- package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -1
- package/dist/src/eval-workflows/evaluation-preparation.js +8 -4
- package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -1
- package/dist/src/eval-workflows/run-evaluation.d.ts +20 -17
- package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
- package/dist/src/eval-workflows/run-evaluation.js +25 -15
- package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
- package/dist/src/executors/claude-cli.d.ts.map +1 -1
- package/dist/src/executors/claude-cli.js +11 -6
- package/dist/src/executors/claude-cli.js.map +1 -1
- package/dist/src/executors/codex-cli-trace.d.ts +10 -0
- package/dist/src/executors/codex-cli-trace.d.ts.map +1 -0
- package/dist/src/executors/codex-cli-trace.js +123 -0
- package/dist/src/executors/codex-cli-trace.js.map +1 -0
- package/dist/src/executors/codex-cli.d.ts +18 -0
- package/dist/src/executors/codex-cli.d.ts.map +1 -0
- package/dist/src/executors/codex-cli.js +254 -0
- package/dist/src/executors/codex-cli.js.map +1 -0
- package/dist/src/executors/codex-sdk.d.ts +18 -0
- package/dist/src/executors/codex-sdk.d.ts.map +1 -0
- package/dist/src/executors/codex-sdk.js +214 -0
- package/dist/src/executors/codex-sdk.js.map +1 -0
- package/dist/src/executors/gemini.d.ts.map +1 -1
- package/dist/src/executors/gemini.js +28 -24
- package/dist/src/executors/gemini.js.map +1 -1
- package/dist/src/executors/index.d.ts.map +1 -1
- package/dist/src/executors/index.js +7 -2
- package/dist/src/executors/index.js.map +1 -1
- package/dist/src/executors/runtime-fingerprint.d.ts +7 -0
- package/dist/src/executors/runtime-fingerprint.d.ts.map +1 -0
- package/dist/src/executors/runtime-fingerprint.js +277 -0
- package/dist/src/executors/runtime-fingerprint.js.map +1 -0
- package/dist/src/executors/script.d.ts.map +1 -1
- package/dist/src/executors/script.js +47 -55
- package/dist/src/executors/script.js.map +1 -1
- package/dist/src/executors/shared.d.ts +78 -1
- package/dist/src/executors/shared.d.ts.map +1 -1
- package/dist/src/executors/shared.js +203 -1
- package/dist/src/executors/shared.js.map +1 -1
- package/dist/src/grading/assertions.d.ts.map +1 -1
- package/dist/src/grading/assertions.js +18 -3
- package/dist/src/grading/assertions.js.map +1 -1
- package/dist/src/grading/index.d.ts.map +1 -1
- package/dist/src/grading/index.js +11 -0
- package/dist/src/grading/index.js.map +1 -1
- package/dist/src/grading/judge.d.ts.map +1 -1
- package/dist/src/grading/judge.js +75 -6
- package/dist/src/grading/judge.js.map +1 -1
- package/dist/src/inputs/skill-loader.d.ts +1 -1
- package/dist/src/inputs/skill-loader.d.ts.map +1 -1
- package/dist/src/inputs/skill-loader.js +1 -1
- package/dist/src/inputs/skill-loader.js.map +1 -1
- package/dist/src/renderer/html-renderer.d.ts +5 -4
- package/dist/src/renderer/html-renderer.d.ts.map +1 -1
- package/dist/src/renderer/html-renderer.js +217 -93
- package/dist/src/renderer/html-renderer.js.map +1 -1
- package/dist/src/renderer/layout.d.ts +2 -1
- package/dist/src/renderer/layout.d.ts.map +1 -1
- package/dist/src/renderer/layout.js +30 -42
- package/dist/src/renderer/layout.js.map +1 -1
- package/dist/src/renderer/summary.d.ts +3 -3
- package/dist/src/renderer/summary.d.ts.map +1 -1
- package/dist/src/renderer/summary.js +232 -42
- package/dist/src/renderer/summary.js.map +1 -1
- package/dist/src/renderer/trends.d.ts.map +1 -1
- package/dist/src/renderer/trends.js +5 -3
- package/dist/src/renderer/trends.js.map +1 -1
- package/dist/src/server/report-server.js +4 -4
- package/dist/src/server/report-server.js.map +1 -1
- package/dist/src/server/report-store.d.ts +7 -5
- package/dist/src/server/report-store.d.ts.map +1 -1
- package/dist/src/server/report-store.js +39 -11
- package/dist/src/server/report-store.js.map +1 -1
- package/dist/src/types/eval.d.ts +4 -2
- package/dist/src/types/eval.d.ts.map +1 -1
- package/dist/src/types/executor.d.ts +5 -0
- package/dist/src/types/executor.d.ts.map +1 -1
- package/dist/src/types/judge.d.ts +10 -0
- package/dist/src/types/judge.d.ts.map +1 -1
- package/dist/src/types/report.d.ts +120 -24
- package/dist/src/types/report.d.ts.map +1 -1
- package/dist/src/types/storage.d.ts +7 -7
- package/dist/src/types/storage.d.ts.map +1 -1
- package/package.json +2 -1
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +0 -153
- package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +0 -1
- package/dist/src/eval-workflows/each-evaluation-workflow.js +0 -178
- package/dist/src/eval-workflows/each-evaluation-workflow.js.map +0 -1
- package/dist/src/executors/openai-cli.d.ts +0 -3
- package/dist/src/executors/openai-cli.d.ts.map +0 -1
- package/dist/src/executors/openai-cli.js +0 -60
- package/dist/src/executors/openai-cli.js.map +0 -1
|
@@ -1,153 +0,0 @@
|
|
|
1
|
-
import type { Artifact, JobStore, ProgressCallback, Report, VarianceData, VariantResult, VariantSummary } from '../types/index.js';
|
|
2
|
-
interface RunSingleEvaluationOptions {
|
|
3
|
-
samplesPath: string;
|
|
4
|
-
skillDir: string;
|
|
5
|
-
artifacts: Artifact[];
|
|
6
|
-
model: string;
|
|
7
|
-
judgeModel: string;
|
|
8
|
-
outputDir: null;
|
|
9
|
-
noJudge: boolean;
|
|
10
|
-
concurrency: number;
|
|
11
|
-
timeoutMs?: number;
|
|
12
|
-
executorName: string;
|
|
13
|
-
judgeExecutorName?: string;
|
|
14
|
-
jobStore: null;
|
|
15
|
-
persistJob: false;
|
|
16
|
-
onProgress: ProgressCallback | null;
|
|
17
|
-
skipPreflight: boolean;
|
|
18
|
-
mcpConfig?: string;
|
|
19
|
-
verbose: boolean;
|
|
20
|
-
/** Forwarded to grade(); each sample × dimension is judged N times. Default 1. */
|
|
21
|
-
judgeRepeat?: number;
|
|
22
|
-
/** Forwarded to pipeline; ≥ 2 entries triggers multi-judge ensemble mode. */
|
|
23
|
-
judgeModels?: import('../types/index.js').JudgeConfig[];
|
|
24
|
-
/** v0.21 Phase 3a length-debias toggle. Default true. */
|
|
25
|
-
lengthDebias?: boolean;
|
|
26
|
-
}
|
|
27
|
-
export interface EachSkillResult {
|
|
28
|
-
name: string;
|
|
29
|
-
artifactHash: string;
|
|
30
|
-
samplesPath: string;
|
|
31
|
-
sampleCount: number;
|
|
32
|
-
summary: Record<string, VariantSummary>;
|
|
33
|
-
/** repeat > 1 时由 runMultiple 聚合,承载三层独立 variance + t 检验 */
|
|
34
|
-
variance?: VarianceData;
|
|
35
|
-
results: Array<{
|
|
36
|
-
sample_id: string;
|
|
37
|
-
variants: {
|
|
38
|
-
baseline: VariantResult;
|
|
39
|
-
skill: VariantResult;
|
|
40
|
-
};
|
|
41
|
-
}>;
|
|
42
|
-
}
|
|
43
|
-
export declare function buildEachReport({ skillDir, skillEntries, skillResults, model, judgeModel, noJudge, executorName, judgeExecutorName, project, owner, tags, dryRun, concurrency, timeoutMs, totalCostUSD, repeat, }: {
|
|
44
|
-
skillDir: string;
|
|
45
|
-
skillEntries: Array<{
|
|
46
|
-
name: string;
|
|
47
|
-
skillPath: string;
|
|
48
|
-
samplesPath: string;
|
|
49
|
-
}>;
|
|
50
|
-
skillResults: EachSkillResult[];
|
|
51
|
-
model: string;
|
|
52
|
-
judgeModel: string;
|
|
53
|
-
noJudge: boolean;
|
|
54
|
-
executorName: string;
|
|
55
|
-
judgeExecutorName?: string;
|
|
56
|
-
project?: string;
|
|
57
|
-
owner?: string;
|
|
58
|
-
tags?: string[];
|
|
59
|
-
dryRun: boolean;
|
|
60
|
-
concurrency: number;
|
|
61
|
-
timeoutMs?: number;
|
|
62
|
-
totalCostUSD: number;
|
|
63
|
-
repeat?: number;
|
|
64
|
-
}): {
|
|
65
|
-
report: {
|
|
66
|
-
id: string;
|
|
67
|
-
each: true;
|
|
68
|
-
meta: {
|
|
69
|
-
variants: string[];
|
|
70
|
-
model: string;
|
|
71
|
-
judgeModel: string | null;
|
|
72
|
-
executor: string;
|
|
73
|
-
sampleCount: number;
|
|
74
|
-
taskCount: number;
|
|
75
|
-
totalCostUSD: number;
|
|
76
|
-
timestamp: string;
|
|
77
|
-
cliVersion: string;
|
|
78
|
-
nodeVersion: string;
|
|
79
|
-
artifactHashes: {
|
|
80
|
-
[k: string]: string;
|
|
81
|
-
};
|
|
82
|
-
request: import("../types/eval.js").EvaluationRequest;
|
|
83
|
-
run: import("../types/eval.js").EvaluationRun;
|
|
84
|
-
job: import("../types/eval.js").EvaluationJob;
|
|
85
|
-
};
|
|
86
|
-
summary: {};
|
|
87
|
-
results: never[];
|
|
88
|
-
overview: {
|
|
89
|
-
totalArtifacts: number;
|
|
90
|
-
totalSamples: number;
|
|
91
|
-
totalCostUSD: number;
|
|
92
|
-
artifacts: {
|
|
93
|
-
name: string;
|
|
94
|
-
baselineScore: number | null;
|
|
95
|
-
artifactScore: number | null;
|
|
96
|
-
improvement: string;
|
|
97
|
-
}[];
|
|
98
|
-
};
|
|
99
|
-
artifacts: EachSkillResult[];
|
|
100
|
-
};
|
|
101
|
-
job: import("../types/eval.js").EvaluationJob;
|
|
102
|
-
};
|
|
103
|
-
export declare function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir, project, owner, tags, noJudge, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, lengthDebias, strictBaseline, variantAllowedSkills, runSingleEvaluation, }: {
|
|
104
|
-
skillDir: string;
|
|
105
|
-
skillEntries: Array<{
|
|
106
|
-
name: string;
|
|
107
|
-
skillPath: string;
|
|
108
|
-
samplesPath: string;
|
|
109
|
-
}>;
|
|
110
|
-
model: string;
|
|
111
|
-
judgeModel: string;
|
|
112
|
-
outputDir?: string | null;
|
|
113
|
-
project?: string;
|
|
114
|
-
owner?: string;
|
|
115
|
-
tags?: string[];
|
|
116
|
-
noJudge?: boolean;
|
|
117
|
-
concurrency?: number;
|
|
118
|
-
timeoutMs?: number;
|
|
119
|
-
executorName: string;
|
|
120
|
-
judgeExecutorName?: string;
|
|
121
|
-
jobStore?: JobStore | null;
|
|
122
|
-
persistJob?: boolean;
|
|
123
|
-
onProgress?: ProgressCallback | null;
|
|
124
|
-
onSkillProgress?: ((info: {
|
|
125
|
-
phase: string;
|
|
126
|
-
skill: string;
|
|
127
|
-
current: number;
|
|
128
|
-
total: number;
|
|
129
|
-
}) => void) | null;
|
|
130
|
-
skipPreflight?: boolean;
|
|
131
|
-
mcpConfig?: string;
|
|
132
|
-
verbose?: boolean;
|
|
133
|
-
repeat?: number;
|
|
134
|
-
judgeRepeat?: number;
|
|
135
|
-
judgeModels?: import('../types/index.js').JudgeConfig[];
|
|
136
|
-
lengthDebias?: boolean;
|
|
137
|
-
/** — strict-baseline default. Forwarded to per-skill resolveArtifacts so each
|
|
138
|
-
* skill's baseline gets allowedSkills=[] when default true. */
|
|
139
|
-
strictBaseline?: boolean;
|
|
140
|
-
/** — explicit per-variant allowedSkills override. each mode rarely uses this
|
|
141
|
-
* (since it auto-pairs baseline + skill), but if user has yaml override on either
|
|
142
|
-
* "baseline" or "skill" name, we honor it. */
|
|
143
|
-
variantAllowedSkills?: Record<string, string[]>;
|
|
144
|
-
runSingleEvaluation: (options: RunSingleEvaluationOptions) => Promise<{
|
|
145
|
-
report: Report;
|
|
146
|
-
filePath: string | null;
|
|
147
|
-
}>;
|
|
148
|
-
}): Promise<{
|
|
149
|
-
report: Report;
|
|
150
|
-
filePath: string | null;
|
|
151
|
-
}>;
|
|
152
|
-
export {};
|
|
153
|
-
//# sourceMappingURL=each-evaluation-workflow.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"each-evaluation-workflow.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,gBAAgB,EAAE,MAAM,EAAE,YAAY,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAEnI,UAAU,0BAA0B;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,IAAI,CAAC;IAChB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,EAAE,IAAI,CAAC;IACf,UAAU,EAAE,KAAK,CAAC;IAClB,UAAU,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACpC,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,kFAAkF;IAClF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,yDAAyD;IACzD,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACxC,0DAA0D;IAC1D,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,OAAO,EAAE,KAAK,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;QAClB,QAAQ,EAAE;YACR,QAAQ,EAAE,aAAa,CAAC;YACxB,KAAK,EAAE,aAAa,CAAC;SACtB,CAAC;KACH,CAAC,CAAC;CACJ;AAsBD,wBAAgB,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,EAAE,OAAO,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAwEA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,WAAe,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,cAAc,EACd,oBAAoB,EACpB,mBAAmB,GACpB,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,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,EAAE,MAAM,CAAC;IACrB,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,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC5G,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;oEACgE;IAChE,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB;;mDAE+C;IAC/C,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;IAChD,mBAAmB,EAAE,CAAC,OAAO,EAAE,0BAA0B,KAAK,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC,CAAC;CACpH,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA0FvD"}
|
|
@@ -1,178 +0,0 @@
|
|
|
1
|
-
import { resolve } from 'node:path';
|
|
2
|
-
import { DEFAULT_OUTPUT_DIR, generateRunId, persistReport } from '../eval-core/evaluation-reporting.js';
|
|
3
|
-
import { buildEvaluationRequest, createEvaluationRun, createSucceededJob, finalizeEvaluationRun } from '../eval-core/evaluation-job.js';
|
|
4
|
-
import { createFileJobStore, DEFAULT_JOBS_DIR } from '../server/job-store.js';
|
|
5
|
-
import { resolveArtifacts } from '../inputs/skill-loader.js';
|
|
6
|
-
function buildEachOverview(skillResults, totalCostUSD) {
|
|
7
|
-
return {
|
|
8
|
-
totalArtifacts: skillResults.length,
|
|
9
|
-
totalSamples: skillResults.reduce((sum, skill) => sum + skill.sampleCount, 0),
|
|
10
|
-
totalCostUSD: Number(totalCostUSD.toFixed(6)),
|
|
11
|
-
artifacts: skillResults.map((skill) => {
|
|
12
|
-
const baselineSummary = skill.summary.baseline;
|
|
13
|
-
const artifactSummary = skill.summary.skill;
|
|
14
|
-
const baselineScore = baselineSummary?.avgCompositeScore ?? baselineSummary?.avgLlmScore ?? null;
|
|
15
|
-
const artifactScore = artifactSummary?.avgCompositeScore ?? artifactSummary?.avgLlmScore ?? null;
|
|
16
|
-
if (typeof baselineScore !== 'number' || typeof artifactScore !== 'number' || baselineScore <= 0) {
|
|
17
|
-
return { name: skill.name, baselineScore, artifactScore, improvement: '-' };
|
|
18
|
-
}
|
|
19
|
-
const delta = ((artifactScore - baselineScore) / baselineScore * 100).toFixed(0);
|
|
20
|
-
const improvement = artifactScore >= baselineScore ? `+${delta}%` : `${delta}%`;
|
|
21
|
-
return { name: skill.name, baselineScore, artifactScore, improvement };
|
|
22
|
-
}),
|
|
23
|
-
};
|
|
24
|
-
}
|
|
25
|
-
export function buildEachReport({ skillDir, skillEntries, skillResults, model, judgeModel, noJudge, executorName, judgeExecutorName, project, owner, tags, dryRun, concurrency, timeoutMs, totalCostUSD, repeat, }) {
|
|
26
|
-
const runId = generateRunId(['each']);
|
|
27
|
-
const request = buildEvaluationRequest({
|
|
28
|
-
samplesPath: '',
|
|
29
|
-
skillDir,
|
|
30
|
-
artifacts: skillEntries.map((entry) => ({
|
|
31
|
-
name: entry.name,
|
|
32
|
-
kind: 'skill',
|
|
33
|
-
source: 'file-path',
|
|
34
|
-
content: null,
|
|
35
|
-
locator: entry.skillPath,
|
|
36
|
-
})),
|
|
37
|
-
model,
|
|
38
|
-
judgeModel: noJudge ? null : judgeModel,
|
|
39
|
-
executor: executorName,
|
|
40
|
-
judgeExecutor: judgeExecutorName || executorName,
|
|
41
|
-
noJudge,
|
|
42
|
-
concurrency,
|
|
43
|
-
timeoutMs,
|
|
44
|
-
noCache: false,
|
|
45
|
-
dryRun,
|
|
46
|
-
blind: false,
|
|
47
|
-
project,
|
|
48
|
-
owner,
|
|
49
|
-
tags,
|
|
50
|
-
repeat,
|
|
51
|
-
each: true,
|
|
52
|
-
});
|
|
53
|
-
const createdAt = new Date().toISOString();
|
|
54
|
-
const { run: initialRun, startedAt } = createEvaluationRun(runId, createdAt);
|
|
55
|
-
const finishedAt = new Date().toISOString();
|
|
56
|
-
const run = finalizeEvaluationRun(initialRun, finishedAt);
|
|
57
|
-
const job = createSucceededJob({
|
|
58
|
-
jobId: `job-${runId}`,
|
|
59
|
-
runId,
|
|
60
|
-
reportId: runId,
|
|
61
|
-
request,
|
|
62
|
-
createdAt,
|
|
63
|
-
startedAt,
|
|
64
|
-
finishedAt,
|
|
65
|
-
});
|
|
66
|
-
const totalSampleCount = skillResults.reduce((sum, skill) => sum + skill.sampleCount, 0);
|
|
67
|
-
return {
|
|
68
|
-
report: {
|
|
69
|
-
id: runId,
|
|
70
|
-
each: true,
|
|
71
|
-
meta: {
|
|
72
|
-
variants: skillEntries.map((entry) => entry.name),
|
|
73
|
-
model,
|
|
74
|
-
judgeModel: noJudge ? null : judgeModel,
|
|
75
|
-
executor: executorName,
|
|
76
|
-
sampleCount: totalSampleCount,
|
|
77
|
-
taskCount: totalSampleCount * 2,
|
|
78
|
-
totalCostUSD: Number(totalCostUSD.toFixed(6)),
|
|
79
|
-
timestamp: new Date().toISOString(),
|
|
80
|
-
cliVersion: '',
|
|
81
|
-
nodeVersion: process.version,
|
|
82
|
-
artifactHashes: Object.fromEntries(skillResults.map((skill) => [skill.name, skill.artifactHash || 'no-skill'])),
|
|
83
|
-
request,
|
|
84
|
-
run,
|
|
85
|
-
job,
|
|
86
|
-
},
|
|
87
|
-
summary: {},
|
|
88
|
-
results: [],
|
|
89
|
-
overview: buildEachOverview(skillResults, totalCostUSD),
|
|
90
|
-
artifacts: skillResults,
|
|
91
|
-
},
|
|
92
|
-
job,
|
|
93
|
-
};
|
|
94
|
-
}
|
|
95
|
-
export async function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, noJudge = false, concurrency = 1, timeoutMs, executorName, judgeExecutorName, jobStore = null, persistJob = true, onProgress = null, onSkillProgress = null, skipPreflight = false, mcpConfig, verbose = false, repeat, judgeRepeat, judgeModels, lengthDebias, strictBaseline, variantAllowedSkills, runSingleEvaluation, }) {
|
|
96
|
-
const skillResults = [];
|
|
97
|
-
let totalCostUSD = 0;
|
|
98
|
-
for (let i = 0; i < skillEntries.length; i++) {
|
|
99
|
-
const entry = skillEntries[i];
|
|
100
|
-
onSkillProgress?.({ phase: 'start', skill: entry.name, current: i + 1, total: skillEntries.length });
|
|
101
|
-
// each mode 的实验结构固定为"baseline (control) vs skill (treatment)"。
|
|
102
|
-
// 显式在 artifact 上填 experimentRole,下游 buildVariantConfig 直接读取。
|
|
103
|
-
// — pass strictBaseline + variantAllowedSkills opts so isolation 默认生效。
|
|
104
|
-
const skillArtifacts = resolveArtifacts(resolve(skillDir), ['baseline', entry.skillPath], { strictBaseline, variantAllowedSkills }).map((artifact) => {
|
|
105
|
-
if (artifact.name === entry.skillPath) {
|
|
106
|
-
return { ...artifact, name: 'skill', experimentRole: 'treatment' };
|
|
107
|
-
}
|
|
108
|
-
return { ...artifact, experimentRole: 'control' };
|
|
109
|
-
});
|
|
110
|
-
const { report } = await runSingleEvaluation({
|
|
111
|
-
samplesPath: entry.samplesPath,
|
|
112
|
-
skillDir,
|
|
113
|
-
artifacts: skillArtifacts,
|
|
114
|
-
model,
|
|
115
|
-
judgeModel,
|
|
116
|
-
outputDir: null,
|
|
117
|
-
noJudge,
|
|
118
|
-
concurrency,
|
|
119
|
-
timeoutMs,
|
|
120
|
-
executorName,
|
|
121
|
-
judgeExecutorName,
|
|
122
|
-
jobStore: null,
|
|
123
|
-
persistJob: false,
|
|
124
|
-
onProgress,
|
|
125
|
-
skipPreflight: skipPreflight || i > 0,
|
|
126
|
-
mcpConfig,
|
|
127
|
-
verbose,
|
|
128
|
-
judgeRepeat,
|
|
129
|
-
judgeModels,
|
|
130
|
-
lengthDebias,
|
|
131
|
-
});
|
|
132
|
-
skillResults.push({
|
|
133
|
-
name: entry.name,
|
|
134
|
-
artifactHash: report.meta.artifactHashes?.skill || '',
|
|
135
|
-
samplesPath: entry.samplesPath,
|
|
136
|
-
sampleCount: report.meta.sampleCount,
|
|
137
|
-
summary: {
|
|
138
|
-
baseline: report.summary.baseline || {},
|
|
139
|
-
skill: (report.summary.skill || {}),
|
|
140
|
-
},
|
|
141
|
-
// runMultiple 跑 N 次后会把 variance 挂到 report.variance,这里搬到 skill 维度
|
|
142
|
-
...(report.variance ? { variance: report.variance } : {}),
|
|
143
|
-
results: report.results.map((result) => ({
|
|
144
|
-
sample_id: result.sample_id,
|
|
145
|
-
variants: {
|
|
146
|
-
baseline: result.variants.baseline || result.variants['baseline'],
|
|
147
|
-
skill: result.variants.skill,
|
|
148
|
-
},
|
|
149
|
-
})),
|
|
150
|
-
});
|
|
151
|
-
totalCostUSD += report.meta.totalCostUSD;
|
|
152
|
-
onSkillProgress?.({ phase: 'done', skill: entry.name, current: i + 1, total: skillEntries.length });
|
|
153
|
-
}
|
|
154
|
-
const { report: combinedReport, job } = buildEachReport({
|
|
155
|
-
skillDir,
|
|
156
|
-
skillEntries,
|
|
157
|
-
skillResults,
|
|
158
|
-
model,
|
|
159
|
-
judgeModel,
|
|
160
|
-
noJudge,
|
|
161
|
-
executorName,
|
|
162
|
-
judgeExecutorName,
|
|
163
|
-
project,
|
|
164
|
-
owner,
|
|
165
|
-
tags,
|
|
166
|
-
dryRun: false,
|
|
167
|
-
concurrency,
|
|
168
|
-
timeoutMs,
|
|
169
|
-
totalCostUSD,
|
|
170
|
-
repeat,
|
|
171
|
-
});
|
|
172
|
-
const filePath = persistReport(combinedReport, outputDir);
|
|
173
|
-
const resolvedJobStore = persistJob ? (jobStore ?? createFileJobStore(DEFAULT_JOBS_DIR)) : null;
|
|
174
|
-
if (resolvedJobStore)
|
|
175
|
-
await resolvedJobStore.save(job.jobId, job);
|
|
176
|
-
return { report: combinedReport, filePath };
|
|
177
|
-
}
|
|
178
|
-
//# sourceMappingURL=each-evaluation-workflow.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"each-evaluation-workflow.js","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACxG,OAAO,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,gCAAgC,CAAC;AACxI,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AA8C7D,SAAS,iBAAiB,CAAC,YAA+B,EAAE,YAAoB;IAC9E,OAAO;QACL,cAAc,EAAE,YAAY,CAAC,MAAM;QACnC,YAAY,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;QAC7E,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7C,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;YACpC,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC;YAC/C,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;YAC5C,MAAM,aAAa,GAAG,eAAe,EAAE,iBAAiB,IAAI,eAAe,EAAE,WAAW,IAAI,IAAI,CAAC;YACjG,MAAM,aAAa,GAAG,eAAe,EAAE,iBAAiB,IAAI,eAAe,EAAE,WAAW,IAAI,IAAI,CAAC;YACjG,IAAI,OAAO,aAAa,KAAK,QAAQ,IAAI,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,IAAI,CAAC,EAAE,CAAC;gBACjG,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,EAAE,aAAa,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC;YAC9E,CAAC;YACD,MAAM,KAAK,GAAG,CAAC,CAAC,aAAa,GAAG,aAAa,CAAC,GAAG,aAAa,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YACjF,MAAM,WAAW,GAAG,aAAa,IAAI,aAAa,CAAC,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC;YAChF,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,aAAa,EAAE,aAAa,EAAE,WAAW,EAAE,CAAC;QACzE,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,GAkBP;IACC,MAAM,KAAK,GAAG,aAAa,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;IACtC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW,EAAE,EAAE;QACf,QAAQ;QACR,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YACtC,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,IAAI,EAAE,OAAO;YACb,MAAM,EAAE,WAAW;YACnB,OAAO,EAAE,IAAI;YACb,OAAO,EAAE,KAAK,CAAC,SAAS;SACzB,CAAC,CAAC;QACH,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB,IAAI,YAAY;QAChD,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO,EAAE,KAAK;QACd,MAAM;QACN,KAAK,EAAE,KAAK;QACZ,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,IAAI,EAAE,IAAI;KACX,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,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAC1D,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,OAAO,KAAK,EAAE;QACrB,KAAK;QACL,QAAQ,EAAE,KAAK;QACf,OAAO;QACP,SAAS;QACT,SAAS;QACT,UAAU;KACX,CAAC,CAAC;IACH,MAAM,gBAAgB,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;IAEzF,OAAO;QACL,MAAM,EAAE;YACN,EAAE,EAAE,KAAK;YACT,IAAI,EAAE,IAAI;YACV,IAAI,EAAE;gBACJ,QAAQ,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC;gBACjD,KAAK;gBACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;gBACvC,QAAQ,EAAE,YAAY;gBACtB,WAAW,EAAE,gBAAgB;gBAC7B,SAAS,EAAE,gBAAgB,GAAG,CAAC;gBAC/B,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;gBAC7C,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;gBACnC,UAAU,EAAE,EAAE;gBACd,WAAW,EAAE,OAAO,CAAC,OAAO;gBAC5B,cAAc,EAAE,MAAM,CAAC,WAAW,CAChC,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,YAAY,IAAI,UAAU,CAAC,CAAC,CAC5E;gBACD,OAAO;gBACP,GAAG;gBACH,GAAG;aACJ;YACD,OAAO,EAAE,EAAE;YACX,OAAO,EAAE,EAAE;YACX,QAAQ,EAAE,iBAAiB,CAAC,YAAY,EAAE,YAAY,CAAC;YACvD,SAAS,EAAE,YAAY;SACP;QAClB,GAAG;KACJ,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,WAAW,GAAG,CAAC,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,eAAe,GAAG,IAAI,EACtB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,cAAc,EACd,oBAAoB,EACpB,mBAAmB,GAkCpB;IACC,MAAM,YAAY,GAAsB,EAAE,CAAC;IAC3C,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,YAAY,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QAC9B,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;QAErG,+DAA+D;QAC/D,6DAA6D;QAC7D,wEAAwE;QACxE,MAAM,cAAc,GAAG,gBAAgB,CACrC,OAAO,CAAC,QAAQ,CAAC,EACjB,CAAC,UAAU,EAAE,KAAK,CAAC,SAAS,CAAC,EAC7B,EAAE,cAAc,EAAE,oBAAoB,EAAE,CACzC,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE;YACjB,IAAI,QAAQ,CAAC,IAAI,KAAK,KAAK,CAAC,SAAS,EAAE,CAAC;gBACtC,OAAO,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE,cAAc,EAAE,WAAoB,EAAE,CAAC;YAC9E,CAAC;YACD,OAAO,EAAE,GAAG,QAAQ,EAAE,cAAc,EAAE,SAAkB,EAAE,CAAC;QAC7D,CAAC,CAAC,CAAC;QACH,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,mBAAmB,CAAC;YAC3C,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,QAAQ;YACR,SAAS,EAAE,cAAc;YACzB,KAAK;YACL,UAAU;YACV,SAAS,EAAE,IAAI;YACf,OAAO;YACP,WAAW;YACX,SAAS;YACT,YAAY;YACZ,iBAAiB;YACjB,QAAQ,EAAE,IAAI;YACd,UAAU,EAAE,KAAK;YACjB,UAAU;YACV,aAAa,EAAE,aAAa,IAAI,CAAC,GAAG,CAAC;YACrC,SAAS;YACT,OAAO;YACP,WAAW;YACX,WAAW;YACX,YAAY;SACb,CAAC,CAAC;QAEH,YAAY,CAAC,IAAI,CAAC;YAChB,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,YAAY,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE,KAAK,IAAI,EAAE;YACrD,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW;YACpC,OAAO,EAAE;gBACP,QAAQ,EAAE,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAK,EAAqB;gBAC3D,KAAK,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,IAAI,EAAE,CAAmB;aACtD;YACD,iEAAiE;YACjE,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACzD,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;gBACvC,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,QAAQ,EAAE;oBACR,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC;oBACjE,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,KAAK;iBAC7B;aACF,CAAC,CAAC;SACJ,CAAC,CAAC;QAEH,YAAY,IAAI,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC;QACzC,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;IACtG,CAAC;IAED,MAAM,EAAE,MAAM,EAAE,cAAc,EAAE,GAAG,EAAE,GAAG,eAAe,CAAC;QACtD,QAAQ;QACR,YAAY;QACZ,YAAY;QACZ,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM,EAAE,KAAK;QACb,WAAW;QACX,SAAS;QACT,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IACH,MAAM,QAAQ,GAAG,aAAa,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IAC1D,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAClE,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,QAAQ,EAAE,CAAC;AAC9C,CAAC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"openai-cli.d.ts","sourceRoot":"","sources":["../../../src/executors/openai-cli.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAYnE,wBAAsB,iBAAiB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,SAA8B,EAAE,EAAE,aAAa,GAAG,OAAO,CAAC,UAAU,CAAC,CAqDrI"}
|
|
@@ -1,60 +0,0 @@
|
|
|
1
|
-
import { asErrorLike, DEFAULT_TIMEOUT_MS, errorMessage, execFileAsync, MAX_BUFFER, parseJson, timeoutExecResult, } from './shared.js';
|
|
2
|
-
export async function openAiCliExecutor({ model, system, prompt, timeoutMs = DEFAULT_TIMEOUT_MS }) {
|
|
3
|
-
const args = ['api', 'chat.completions.create', '-m', model || 'gpt-4o'];
|
|
4
|
-
if (system) {
|
|
5
|
-
args.push('-g', 'system', system);
|
|
6
|
-
}
|
|
7
|
-
args.push('-g', 'user', prompt);
|
|
8
|
-
const start = Date.now();
|
|
9
|
-
try {
|
|
10
|
-
const { stdout } = await execFileAsync('openai', args, {
|
|
11
|
-
maxBuffer: MAX_BUFFER,
|
|
12
|
-
timeout: timeoutMs,
|
|
13
|
-
env: { ...process.env },
|
|
14
|
-
});
|
|
15
|
-
const durationMs = Date.now() - start;
|
|
16
|
-
const data = parseJson(stdout);
|
|
17
|
-
const usage = data.usage || {};
|
|
18
|
-
const content = data.choices?.[0]?.message?.content || '';
|
|
19
|
-
const finishReason = data.choices?.[0]?.finish_reason || 'unknown';
|
|
20
|
-
return {
|
|
21
|
-
ok: true,
|
|
22
|
-
durationMs,
|
|
23
|
-
durationApiMs: 0,
|
|
24
|
-
inputTokens: usage.prompt_tokens || 0,
|
|
25
|
-
outputTokens: usage.completion_tokens || 0,
|
|
26
|
-
cacheReadTokens: usage.prompt_tokens_details?.cached_tokens || 0,
|
|
27
|
-
cacheCreationTokens: 0,
|
|
28
|
-
costUSD: 0,
|
|
29
|
-
output: content,
|
|
30
|
-
stopReason: finishReason,
|
|
31
|
-
numTurns: 1,
|
|
32
|
-
};
|
|
33
|
-
}
|
|
34
|
-
catch (err) {
|
|
35
|
-
const durationMs = Date.now() - start;
|
|
36
|
-
const details = asErrorLike(err);
|
|
37
|
-
if (details.killed)
|
|
38
|
-
return timeoutExecResult(timeoutMs, durationMs);
|
|
39
|
-
let parsed = null;
|
|
40
|
-
try {
|
|
41
|
-
parsed = parseJson(details.stdout || '');
|
|
42
|
-
}
|
|
43
|
-
catch { }
|
|
44
|
-
return {
|
|
45
|
-
ok: false,
|
|
46
|
-
error: parsed?.error?.message || errorMessage(err),
|
|
47
|
-
durationMs,
|
|
48
|
-
durationApiMs: 0,
|
|
49
|
-
inputTokens: 0,
|
|
50
|
-
outputTokens: 0,
|
|
51
|
-
cacheReadTokens: 0,
|
|
52
|
-
cacheCreationTokens: 0,
|
|
53
|
-
costUSD: 0,
|
|
54
|
-
output: null,
|
|
55
|
-
stopReason: 'error',
|
|
56
|
-
numTurns: 0,
|
|
57
|
-
};
|
|
58
|
-
}
|
|
59
|
-
}
|
|
60
|
-
//# sourceMappingURL=openai-cli.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"openai-cli.js","sourceRoot":"","sources":["../../../src/executors/openai-cli.ts"],"names":[],"mappings":"AACA,OAAO,EACL,WAAW,EACX,kBAAkB,EAClB,YAAY,EACZ,aAAa,EACb,UAAU,EAEV,SAAS,EACT,iBAAiB,GAClB,MAAM,aAAa,CAAC;AAErB,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,SAAS,GAAG,kBAAkB,EAAiB;IAC9G,MAAM,IAAI,GAAG,CAAC,KAAK,EAAE,yBAAyB,EAAE,IAAI,EAAE,KAAK,IAAI,QAAQ,CAAC,CAAC;IACzE,IAAI,MAAM,EAAE,CAAC;QACX,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;IACpC,CAAC;IACD,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;IAEhC,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,IAAI,CAAC;QACH,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,aAAa,CAAC,QAAQ,EAAE,IAAI,EAAE;YACrD,SAAS,EAAE,UAAU;YACrB,OAAO,EAAE,SAAS;YAClB,GAAG,EAAE,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE;SACxB,CAAC,CAAC;QACH,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;QACtC,MAAM,IAAI,GAAG,SAAS,CAAiB,MAAM,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC;QAC/B,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,IAAI,EAAE,CAAC;QAC1D,MAAM,YAAY,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,aAAa,IAAI,SAAS,CAAC;QACnE,OAAO;YACL,EAAE,EAAE,IAAI;YACR,UAAU;YACV,aAAa,EAAE,CAAC;YAChB,WAAW,EAAE,KAAK,CAAC,aAAa,IAAI,CAAC;YACrC,YAAY,EAAE,KAAK,CAAC,iBAAiB,IAAI,CAAC;YAC1C,eAAe,EAAE,KAAK,CAAC,qBAAqB,EAAE,aAAa,IAAI,CAAC;YAChE,mBAAmB,EAAE,CAAC;YACtB,OAAO,EAAE,CAAC;YACV,MAAM,EAAE,OAAO;YACf,UAAU,EAAE,YAAY;YACxB,QAAQ,EAAE,CAAC;SACZ,CAAC;IACJ,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;QACtC,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QACjC,IAAI,OAAO,CAAC,MAAM;YAAE,OAAO,iBAAiB,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;QACpE,IAAI,MAAM,GAA0B,IAAI,CAAC;QACzC,IAAI,CAAC;YAAC,MAAM,GAAG,SAAS,CAAiB,OAAO,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC;QAC1E,OAAO;YACL,EAAE,EAAE,KAAK;YACT,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,IAAI,YAAY,CAAC,GAAG,CAAC;YAClD,UAAU;YACV,aAAa,EAAE,CAAC;YAChB,WAAW,EAAE,CAAC;YACd,YAAY,EAAE,CAAC;YACf,eAAe,EAAE,CAAC;YAClB,mBAAmB,EAAE,CAAC;YACtB,OAAO,EAAE,CAAC;YACV,MAAM,EAAE,IAAI;YACZ,UAAU,EAAE,OAAO;YACnB,QAAQ,EAAE,CAAC;SACZ,CAAC;IACJ,CAAC;AACH,CAAC"}
|