@llm4ts/flow 0.1.0
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 +41 -0
- package/dist/AzureDevOpsTool.d.ts +60 -0
- package/dist/AzureDevOpsTool.d.ts.map +1 -0
- package/dist/AzureDevOpsTool.js +163 -0
- package/dist/AzureDevOpsTool.js.map +1 -0
- package/dist/Bench.d.ts +133 -0
- package/dist/Bench.d.ts.map +1 -0
- package/dist/Bench.js +272 -0
- package/dist/Bench.js.map +1 -0
- package/dist/BenchReport.d.ts +9 -0
- package/dist/BenchReport.d.ts.map +1 -0
- package/dist/BenchReport.js +360 -0
- package/dist/BenchReport.js.map +1 -0
- package/dist/CapabilityGuard.d.ts +6 -0
- package/dist/CapabilityGuard.d.ts.map +1 -0
- package/dist/CapabilityGuard.js +25 -0
- package/dist/CapabilityGuard.js.map +1 -0
- package/dist/Chat.d.ts +15 -0
- package/dist/Chat.d.ts.map +1 -0
- package/dist/Chat.js +36 -0
- package/dist/Chat.js.map +1 -0
- package/dist/Classified.d.ts +14 -0
- package/dist/Classified.d.ts.map +1 -0
- package/dist/Classified.js +48 -0
- package/dist/Classified.js.map +1 -0
- package/dist/CostLedger.d.ts +55 -0
- package/dist/CostLedger.d.ts.map +1 -0
- package/dist/CostLedger.js +126 -0
- package/dist/CostLedger.js.map +1 -0
- package/dist/CostTracker.d.ts +13 -0
- package/dist/CostTracker.d.ts.map +1 -0
- package/dist/CostTracker.js +116 -0
- package/dist/CostTracker.js.map +1 -0
- package/dist/Equiv.d.ts +163 -0
- package/dist/Equiv.d.ts.map +1 -0
- package/dist/Equiv.js +256 -0
- package/dist/Equiv.js.map +1 -0
- package/dist/EquivReport.d.ts +16 -0
- package/dist/EquivReport.d.ts.map +1 -0
- package/dist/EquivReport.js +77 -0
- package/dist/EquivReport.js.map +1 -0
- package/dist/Flow.d.ts +25 -0
- package/dist/Flow.d.ts.map +1 -0
- package/dist/Flow.js +39 -0
- package/dist/Flow.js.map +1 -0
- package/dist/FlowContext.d.ts +23 -0
- package/dist/FlowContext.d.ts.map +1 -0
- package/dist/FlowContext.js +4 -0
- package/dist/FlowContext.js.map +1 -0
- package/dist/FlowError.d.ts +80 -0
- package/dist/FlowError.d.ts.map +1 -0
- package/dist/FlowError.js +92 -0
- package/dist/FlowError.js.map +1 -0
- package/dist/FlowEvents.d.ts +101 -0
- package/dist/FlowEvents.d.ts.map +1 -0
- package/dist/FlowEvents.js +103 -0
- package/dist/FlowEvents.js.map +1 -0
- package/dist/FlowRecorder.d.ts +26 -0
- package/dist/FlowRecorder.d.ts.map +1 -0
- package/dist/FlowRecorder.js +74 -0
- package/dist/FlowRecorder.js.map +1 -0
- package/dist/GitHubTool.d.ts +53 -0
- package/dist/GitHubTool.d.ts.map +1 -0
- package/dist/GitHubTool.js +191 -0
- package/dist/GitHubTool.js.map +1 -0
- package/dist/GitTool.d.ts +46 -0
- package/dist/GitTool.d.ts.map +1 -0
- package/dist/GitTool.js +97 -0
- package/dist/GitTool.js.map +1 -0
- package/dist/Interaction.d.ts +21 -0
- package/dist/Interaction.d.ts.map +1 -0
- package/dist/Interaction.js +9 -0
- package/dist/Interaction.js.map +1 -0
- package/dist/McpServer.d.ts +18 -0
- package/dist/McpServer.d.ts.map +1 -0
- package/dist/McpServer.js +122 -0
- package/dist/McpServer.js.map +1 -0
- package/dist/Mermaid.d.ts +7 -0
- package/dist/Mermaid.d.ts.map +1 -0
- package/dist/Mermaid.js +98 -0
- package/dist/Mermaid.js.map +1 -0
- package/dist/Pack.d.ts +40 -0
- package/dist/Pack.d.ts.map +1 -0
- package/dist/Pack.js +141 -0
- package/dist/Pack.js.map +1 -0
- package/dist/Package.d.ts +2 -0
- package/dist/Package.d.ts.map +1 -0
- package/dist/Package.js +2 -0
- package/dist/Package.js.map +1 -0
- package/dist/Persistence.d.ts +32 -0
- package/dist/Persistence.d.ts.map +1 -0
- package/dist/Persistence.js +85 -0
- package/dist/Persistence.js.map +1 -0
- package/dist/Plan.d.ts +26 -0
- package/dist/Plan.d.ts.map +1 -0
- package/dist/Plan.js +101 -0
- package/dist/Plan.js.map +1 -0
- package/dist/PlanExecution.d.ts +7 -0
- package/dist/PlanExecution.d.ts.map +1 -0
- package/dist/PlanExecution.js +19 -0
- package/dist/PlanExecution.js.map +1 -0
- package/dist/Planner.d.ts +32 -0
- package/dist/Planner.d.ts.map +1 -0
- package/dist/Planner.js +97 -0
- package/dist/Planner.js.map +1 -0
- package/dist/PrSummary.d.ts +15 -0
- package/dist/PrSummary.d.ts.map +1 -0
- package/dist/PrSummary.js +29 -0
- package/dist/PrSummary.js.map +1 -0
- package/dist/PriceList.d.ts +11 -0
- package/dist/PriceList.d.ts.map +1 -0
- package/dist/PriceList.js +79 -0
- package/dist/PriceList.js.map +1 -0
- package/dist/Provenance.d.ts +27 -0
- package/dist/Provenance.d.ts.map +1 -0
- package/dist/Provenance.js +38 -0
- package/dist/Provenance.js.map +1 -0
- package/dist/Replay.d.ts +29 -0
- package/dist/Replay.d.ts.map +1 -0
- package/dist/Replay.js +181 -0
- package/dist/Replay.js.map +1 -0
- package/dist/Review.d.ts +71 -0
- package/dist/Review.d.ts.map +1 -0
- package/dist/Review.js +204 -0
- package/dist/Review.js.map +1 -0
- package/dist/ReviewCache.d.ts +6 -0
- package/dist/ReviewCache.d.ts.map +1 -0
- package/dist/ReviewCache.js +28 -0
- package/dist/ReviewCache.js.map +1 -0
- package/dist/Reviewer.d.ts +12 -0
- package/dist/Reviewer.d.ts.map +1 -0
- package/dist/Reviewer.js +30 -0
- package/dist/Reviewer.js.map +1 -0
- package/dist/SpecChecks.d.ts +17 -0
- package/dist/SpecChecks.d.ts.map +1 -0
- package/dist/SpecChecks.js +107 -0
- package/dist/SpecChecks.js.map +1 -0
- package/dist/Survey.d.ts +32 -0
- package/dist/Survey.d.ts.map +1 -0
- package/dist/Survey.js +119 -0
- package/dist/Survey.js.map +1 -0
- package/dist/TransientRetry.d.ts +16 -0
- package/dist/TransientRetry.d.ts.map +1 -0
- package/dist/TransientRetry.js +102 -0
- package/dist/TransientRetry.js.map +1 -0
- package/dist/Workspace.d.ts +39 -0
- package/dist/Workspace.d.ts.map +1 -0
- package/dist/Workspace.js +192 -0
- package/dist/Workspace.js.map +1 -0
- package/dist/WorkspaceLayout.d.ts +3 -0
- package/dist/WorkspaceLayout.d.ts.map +1 -0
- package/dist/WorkspaceLayout.js +19 -0
- package/dist/WorkspaceLayout.js.map +1 -0
- package/dist/WorkspaceTools.d.ts +4 -0
- package/dist/WorkspaceTools.d.ts.map +1 -0
- package/dist/WorkspaceTools.js +130 -0
- package/dist/WorkspaceTools.js.map +1 -0
- package/package.json +91 -0
- package/src/AzureDevOpsTool.ts +291 -0
- package/src/Bench.ts +340 -0
- package/src/BenchReport.ts +447 -0
- package/src/CapabilityGuard.ts +43 -0
- package/src/Chat.ts +60 -0
- package/src/Classified.ts +67 -0
- package/src/CostLedger.ts +171 -0
- package/src/CostTracker.ts +170 -0
- package/src/Equiv.ts +378 -0
- package/src/EquivReport.ts +92 -0
- package/src/Flow.ts +72 -0
- package/src/FlowContext.ts +23 -0
- package/src/FlowError.ts +112 -0
- package/src/FlowEvents.ts +147 -0
- package/src/FlowRecorder.ts +116 -0
- package/src/GitHubTool.ts +315 -0
- package/src/GitTool.ts +214 -0
- package/src/Interaction.ts +23 -0
- package/src/McpServer.ts +168 -0
- package/src/Mermaid.ts +112 -0
- package/src/Pack.ts +213 -0
- package/src/Package.ts +1 -0
- package/src/Persistence.ts +198 -0
- package/src/Plan.ts +119 -0
- package/src/PlanExecution.ts +46 -0
- package/src/Planner.ts +134 -0
- package/src/PrSummary.ts +37 -0
- package/src/PriceList.ts +91 -0
- package/src/Provenance.ts +85 -0
- package/src/Replay.ts +265 -0
- package/src/Review.ts +337 -0
- package/src/ReviewCache.ts +45 -0
- package/src/Reviewer.ts +31 -0
- package/src/SpecChecks.ts +143 -0
- package/src/Survey.ts +144 -0
- package/src/TransientRetry.ts +238 -0
- package/src/Workspace.ts +255 -0
- package/src/WorkspaceLayout.ts +22 -0
- package/src/WorkspaceTools.ts +188 -0
package/dist/Bench.js
ADDED
|
@@ -0,0 +1,272 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import * as Ref from "effect/Ref";
|
|
3
|
+
import * as Schema from "effect/Schema";
|
|
4
|
+
import * as Stream from "effect/Stream";
|
|
5
|
+
import { TokenUsage } from "@llm4ts/core/Models";
|
|
6
|
+
import { estimateCostUsd } from "./PriceList.js";
|
|
7
|
+
export class BenchTokens extends Schema.Class("BenchTokens")({
|
|
8
|
+
prompt: Schema.Int,
|
|
9
|
+
completion: Schema.Int,
|
|
10
|
+
cached: Schema.Int
|
|
11
|
+
}) {
|
|
12
|
+
get total() {
|
|
13
|
+
return this.prompt + this.completion;
|
|
14
|
+
}
|
|
15
|
+
add(that) {
|
|
16
|
+
return new BenchTokens({
|
|
17
|
+
prompt: this.prompt + that.prompt,
|
|
18
|
+
completion: this.completion + that.completion,
|
|
19
|
+
cached: this.cached + that.cached
|
|
20
|
+
});
|
|
21
|
+
}
|
|
22
|
+
}
|
|
23
|
+
export const emptyBenchTokens = () => BenchTokens.make({ prompt: 0, completion: 0, cached: 0 });
|
|
24
|
+
export const benchTokensFromUsage = (usage) => BenchTokens.make({
|
|
25
|
+
prompt: usage.prompt,
|
|
26
|
+
completion: usage.completion,
|
|
27
|
+
cached: usage.cached ?? 0
|
|
28
|
+
});
|
|
29
|
+
export class BenchCounters extends Schema.Class("BenchCounters")({
|
|
30
|
+
flakyRetries: Schema.Int,
|
|
31
|
+
transientRetries: Schema.Int,
|
|
32
|
+
autoResumes: Schema.Int,
|
|
33
|
+
turnLimitTrips: Schema.Int,
|
|
34
|
+
shrinks: Schema.Int
|
|
35
|
+
}) {
|
|
36
|
+
get selfHealing() {
|
|
37
|
+
return this.flakyRetries + this.transientRetries + this.autoResumes + this.shrinks;
|
|
38
|
+
}
|
|
39
|
+
}
|
|
40
|
+
export const emptyBenchCounters = () => BenchCounters.make({
|
|
41
|
+
flakyRetries: 0,
|
|
42
|
+
transientRetries: 0,
|
|
43
|
+
autoResumes: 0,
|
|
44
|
+
turnLimitTrips: 0,
|
|
45
|
+
shrinks: 0
|
|
46
|
+
});
|
|
47
|
+
export class BenchPhase extends Schema.Class("BenchPhase")({
|
|
48
|
+
name: Schema.String,
|
|
49
|
+
ms: Schema.Int,
|
|
50
|
+
tokens: BenchTokens,
|
|
51
|
+
costUsd: Schema.optionalKey(Schema.Number),
|
|
52
|
+
counters: BenchCounters,
|
|
53
|
+
detail: Schema.Record(Schema.String, Schema.String)
|
|
54
|
+
}) {
|
|
55
|
+
}
|
|
56
|
+
export class BenchScore extends Schema.Class("BenchScore")({
|
|
57
|
+
name: Schema.String,
|
|
58
|
+
score: Schema.Int,
|
|
59
|
+
max: Schema.Int,
|
|
60
|
+
reasoning: Schema.String
|
|
61
|
+
}) {
|
|
62
|
+
}
|
|
63
|
+
export class BenchJudge extends Schema.Class("BenchJudge")({
|
|
64
|
+
provider: Schema.String,
|
|
65
|
+
model: Schema.String,
|
|
66
|
+
fixed: Schema.Boolean
|
|
67
|
+
}) {
|
|
68
|
+
}
|
|
69
|
+
export class BenchMachine extends Schema.Class("BenchMachine")({
|
|
70
|
+
hostname: Schema.String,
|
|
71
|
+
os: Schema.String,
|
|
72
|
+
arch: Schema.String,
|
|
73
|
+
cores: Schema.Int,
|
|
74
|
+
memoryGb: Schema.Number,
|
|
75
|
+
runtime: Schema.String
|
|
76
|
+
}) {
|
|
77
|
+
}
|
|
78
|
+
export class BenchQuality extends Schema.Class("BenchQuality")({
|
|
79
|
+
buildPassed: Schema.optionalKey(Schema.Boolean),
|
|
80
|
+
testsRun: Schema.optionalKey(Schema.Int),
|
|
81
|
+
testsFailed: Schema.optionalKey(Schema.Int),
|
|
82
|
+
coveragePct: Schema.optionalKey(Schema.Number),
|
|
83
|
+
featureFiles: Schema.Int,
|
|
84
|
+
scenarios: Schema.Int,
|
|
85
|
+
malformedFeatures: Schema.Int,
|
|
86
|
+
specFiles: Schema.Int,
|
|
87
|
+
programs: Schema.Int,
|
|
88
|
+
sourceFiles: Schema.Int,
|
|
89
|
+
sourceLines: Schema.Int,
|
|
90
|
+
testFiles: Schema.Int,
|
|
91
|
+
testLines: Schema.Int,
|
|
92
|
+
deterministicFindings: Schema.Int,
|
|
93
|
+
judgeFindings: Schema.Int,
|
|
94
|
+
gateCleared: Schema.optionalKey(Schema.Boolean),
|
|
95
|
+
gateRounds: Schema.optionalKey(Schema.Int),
|
|
96
|
+
gateOpenIssues: Schema.optionalKey(Schema.Int),
|
|
97
|
+
judgeScores: Schema.Array(BenchScore)
|
|
98
|
+
}) {
|
|
99
|
+
}
|
|
100
|
+
export class BenchRecord extends Schema.Class("BenchRecord")({
|
|
101
|
+
schemaVersion: Schema.Int,
|
|
102
|
+
runId: Schema.String,
|
|
103
|
+
label: Schema.optionalKey(Schema.String),
|
|
104
|
+
startedAt: Schema.String,
|
|
105
|
+
finishedAt: Schema.String,
|
|
106
|
+
provider: Schema.String,
|
|
107
|
+
modelRequested: Schema.String,
|
|
108
|
+
modelsServed: Schema.Array(Schema.String),
|
|
109
|
+
judge: BenchJudge,
|
|
110
|
+
toolVersion: Schema.optionalKey(Schema.String),
|
|
111
|
+
llm4tsVersion: Schema.String,
|
|
112
|
+
pack: Schema.String,
|
|
113
|
+
fixtureFingerprint: Schema.String,
|
|
114
|
+
machine: BenchMachine,
|
|
115
|
+
outcome: Schema.String,
|
|
116
|
+
resumed: Schema.optionalKey(Schema.Boolean),
|
|
117
|
+
totalMs: Schema.Int,
|
|
118
|
+
phases: Schema.Array(BenchPhase),
|
|
119
|
+
quality: Schema.optionalKey(BenchQuality)
|
|
120
|
+
}) {
|
|
121
|
+
get completed() {
|
|
122
|
+
return this.outcome === "completed";
|
|
123
|
+
}
|
|
124
|
+
get totalTokens() {
|
|
125
|
+
return this.phases.reduce((sum, phase) => sum + phase.tokens.total + phase.tokens.cached, 0);
|
|
126
|
+
}
|
|
127
|
+
get totalCostUsd() {
|
|
128
|
+
const values = this.phases.flatMap((phase) => phase.costUsd === undefined ? [] : [phase.costUsd]);
|
|
129
|
+
return values.length === 0 ? undefined : values.reduce((sum, value) => sum + value, 0);
|
|
130
|
+
}
|
|
131
|
+
get selfHealing() {
|
|
132
|
+
return this.phases.reduce((sum, phase) => sum + phase.counters.selfHealing, 0);
|
|
133
|
+
}
|
|
134
|
+
}
|
|
135
|
+
export const CurrentBenchSchema = 1;
|
|
136
|
+
export const BenchUnstaged = "(none)";
|
|
137
|
+
export const emptyBenchObservation = () => ({
|
|
138
|
+
openStages: [],
|
|
139
|
+
cells: new Map(),
|
|
140
|
+
cellKeys: new Map(),
|
|
141
|
+
counters: new Map(),
|
|
142
|
+
modelsServed: new Set()
|
|
143
|
+
});
|
|
144
|
+
const currentStage = (observation) => observation.openStages[observation.openStages.length - 1] ?? BenchUnstaged;
|
|
145
|
+
const dropFirst = (values, value) => {
|
|
146
|
+
const index = values.indexOf(value);
|
|
147
|
+
return index < 0 ? values : [...values.slice(0, index), ...values.slice(index + 1)];
|
|
148
|
+
};
|
|
149
|
+
const classifyNotice = (message, counters) => {
|
|
150
|
+
if (message.startsWith("⟳ flaky")) {
|
|
151
|
+
return new BenchCounters({
|
|
152
|
+
...counters,
|
|
153
|
+
flakyRetries: counters.flakyRetries + 1
|
|
154
|
+
});
|
|
155
|
+
}
|
|
156
|
+
if (message.startsWith("⟳ transient")) {
|
|
157
|
+
return new BenchCounters({
|
|
158
|
+
...counters,
|
|
159
|
+
transientRetries: counters.transientRetries + 1
|
|
160
|
+
});
|
|
161
|
+
}
|
|
162
|
+
if (message.startsWith("↻ auto-resume")) {
|
|
163
|
+
return new BenchCounters({
|
|
164
|
+
...counters,
|
|
165
|
+
autoResumes: counters.autoResumes + 1
|
|
166
|
+
});
|
|
167
|
+
}
|
|
168
|
+
if (message.includes("turn limit hit")) {
|
|
169
|
+
return new BenchCounters({
|
|
170
|
+
...counters,
|
|
171
|
+
turnLimitTrips: counters.turnLimitTrips + 1
|
|
172
|
+
});
|
|
173
|
+
}
|
|
174
|
+
if (message.includes("shrinking to")) {
|
|
175
|
+
return new BenchCounters({
|
|
176
|
+
...counters,
|
|
177
|
+
shrinks: counters.shrinks + 1
|
|
178
|
+
});
|
|
179
|
+
}
|
|
180
|
+
return undefined;
|
|
181
|
+
};
|
|
182
|
+
export const observeBench = (observation, event) => {
|
|
183
|
+
switch (event._tag) {
|
|
184
|
+
case "StageStarted":
|
|
185
|
+
return {
|
|
186
|
+
...observation,
|
|
187
|
+
openStages: [event.stage, ...observation.openStages]
|
|
188
|
+
};
|
|
189
|
+
case "StageCompleted":
|
|
190
|
+
case "StageFailed":
|
|
191
|
+
return {
|
|
192
|
+
...observation,
|
|
193
|
+
openStages: dropFirst(observation.openStages, event.stage)
|
|
194
|
+
};
|
|
195
|
+
case "TokensUsed": {
|
|
196
|
+
const stage = currentStage(observation);
|
|
197
|
+
const model = event.model ?? "";
|
|
198
|
+
const key = JSON.stringify([stage, model]);
|
|
199
|
+
const cells = new Map(observation.cells);
|
|
200
|
+
cells.set(key, (cells.get(key) ?? emptyBenchTokens()).add(benchTokensFromUsage(event.usage)));
|
|
201
|
+
return {
|
|
202
|
+
...observation,
|
|
203
|
+
cells,
|
|
204
|
+
cellKeys: new Map(observation.cellKeys).set(key, [stage, model]),
|
|
205
|
+
modelsServed: event.model === undefined
|
|
206
|
+
? observation.modelsServed
|
|
207
|
+
: new Set([...observation.modelsServed, event.model])
|
|
208
|
+
};
|
|
209
|
+
}
|
|
210
|
+
case "Info": {
|
|
211
|
+
const stage = currentStage(observation);
|
|
212
|
+
const current = observation.counters.get(stage) ?? emptyBenchCounters();
|
|
213
|
+
const next = classifyNotice(event.message, current);
|
|
214
|
+
return next === undefined
|
|
215
|
+
? observation
|
|
216
|
+
: {
|
|
217
|
+
...observation,
|
|
218
|
+
counters: new Map(observation.counters).set(stage, next)
|
|
219
|
+
};
|
|
220
|
+
}
|
|
221
|
+
default:
|
|
222
|
+
return observation;
|
|
223
|
+
}
|
|
224
|
+
};
|
|
225
|
+
export const stageBenchTokens = (observation, stage) => [...observation.cells.entries()].reduce((total, [key, tokens]) => {
|
|
226
|
+
const parts = observation.cellKeys.get(key);
|
|
227
|
+
return parts?.[0] === stage ? total.add(tokens) : total;
|
|
228
|
+
}, emptyBenchTokens());
|
|
229
|
+
export const stageBenchCounters = (observation, stage) => observation.counters.get(stage) ?? emptyBenchCounters();
|
|
230
|
+
export const makeBenchPhase = (observation, stage, ms, detail = {}) => {
|
|
231
|
+
const costs = [...observation.cells.entries()].flatMap(([key, tokens]) => {
|
|
232
|
+
const parts = observation.cellKeys.get(key);
|
|
233
|
+
if (parts?.[0] !== stage) {
|
|
234
|
+
return [];
|
|
235
|
+
}
|
|
236
|
+
const cost = estimateCostUsd(parts[1], TokenUsage.make({
|
|
237
|
+
prompt: tokens.prompt,
|
|
238
|
+
completion: tokens.completion,
|
|
239
|
+
total: tokens.total,
|
|
240
|
+
...(tokens.cached <= 0 ? {} : { cached: tokens.cached })
|
|
241
|
+
}));
|
|
242
|
+
return cost === undefined ? [] : [cost];
|
|
243
|
+
});
|
|
244
|
+
return BenchPhase.make({
|
|
245
|
+
name: stage,
|
|
246
|
+
ms,
|
|
247
|
+
tokens: stageBenchTokens(observation, stage),
|
|
248
|
+
counters: stageBenchCounters(observation, stage),
|
|
249
|
+
detail,
|
|
250
|
+
...(costs.length === 0 ? {} : { costUsd: costs.reduce((sum, cost) => sum + cost, 0) })
|
|
251
|
+
});
|
|
252
|
+
};
|
|
253
|
+
const isHub = (events) => "subscribe" in events && "publishedCount" in events && "stream" in events;
|
|
254
|
+
export const makeBenchTap = Effect.fn("@llm4ts/flow/Bench.makeTap")(function* (events) {
|
|
255
|
+
const observation = yield* Ref.make(emptyBenchObservation());
|
|
256
|
+
const consumed = yield* Ref.make(0);
|
|
257
|
+
if (!isHub(events)) {
|
|
258
|
+
return {
|
|
259
|
+
get: Ref.get(observation),
|
|
260
|
+
reset: Ref.set(observation, emptyBenchObservation())
|
|
261
|
+
};
|
|
262
|
+
}
|
|
263
|
+
const subscription = yield* events.subscribe;
|
|
264
|
+
const baseline = yield* events.publishedCount;
|
|
265
|
+
yield* Stream.fromSubscription(subscription).pipe(Stream.runForEach((event) => Ref.update(observation, (current) => observeBench(current, event)).pipe(Effect.andThen(Ref.update(consumed, (count) => count + 1)))), Effect.forkScoped);
|
|
266
|
+
const drain = Effect.suspend(() => Effect.all([events.publishedCount, Ref.get(consumed)]).pipe(Effect.flatMap(([published, seen]) => seen >= published - baseline ? Effect.void : Effect.yieldNow.pipe(Effect.andThen(drain)))));
|
|
267
|
+
return {
|
|
268
|
+
get: drain.pipe(Effect.andThen(Ref.get(observation))),
|
|
269
|
+
reset: drain.pipe(Effect.andThen(Ref.set(observation, emptyBenchObservation())))
|
|
270
|
+
};
|
|
271
|
+
});
|
|
272
|
+
//# sourceMappingURL=Bench.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Bench.js","sourceRoot":"","sources":["../src/Bench.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,GAAG,MAAM,YAAY,CAAA;AACjC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAEvC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAA;AAEhD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAA;AAEhD,MAAM,OAAO,WAAY,SAAQ,MAAM,CAAC,KAAK,CAAc,aAAa,CAAC,CAAC;IACxE,MAAM,EAAE,MAAM,CAAC,GAAG;IAClB,UAAU,EAAE,MAAM,CAAC,GAAG;IACtB,MAAM,EAAE,MAAM,CAAC,GAAG;CACnB,CAAC;IACA,IAAI,KAAK;QACP,OAAO,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,UAAU,CAAA;IACtC,CAAC;IAED,GAAG,CAAC,IAAiB;QACnB,OAAO,IAAI,WAAW,CAAC;YACrB,MAAM,EAAE,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM;YACjC,UAAU,EAAE,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,UAAU;YAC7C,MAAM,EAAE,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM;SAClC,CAAC,CAAA;IACJ,CAAC;CACF;AAED,MAAM,CAAC,MAAM,gBAAgB,GAAG,GAAgB,EAAE,CAChD,WAAW,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC,CAAA;AAE3D,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,KAAiB,EAAe,EAAE,CACrE,WAAW,CAAC,IAAI,CAAC;IACf,MAAM,EAAE,KAAK,CAAC,MAAM;IACpB,UAAU,EAAE,KAAK,CAAC,UAAU;IAC5B,MAAM,EAAE,KAAK,CAAC,MAAM,IAAI,CAAC;CAC1B,CAAC,CAAA;AAEJ,MAAM,OAAO,aAAc,SAAQ,MAAM,CAAC,KAAK,CAAgB,eAAe,CAAC,CAAC;IAC9E,YAAY,EAAE,MAAM,CAAC,GAAG;IACxB,gBAAgB,EAAE,MAAM,CAAC,GAAG;IAC5B,WAAW,EAAE,MAAM,CAAC,GAAG;IACvB,cAAc,EAAE,MAAM,CAAC,GAAG;IAC1B,OAAO,EAAE,MAAM,CAAC,GAAG;CACpB,CAAC;IACA,IAAI,WAAW;QACb,OAAO,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,OAAO,CAAA;IACpF,CAAC;CACF;AAED,MAAM,CAAC,MAAM,kBAAkB,GAAG,GAAkB,EAAE,CACpD,aAAa,CAAC,IAAI,CAAC;IACjB,YAAY,EAAE,CAAC;IACf,gBAAgB,EAAE,CAAC;IACnB,WAAW,EAAE,CAAC;IACd,cAAc,EAAE,CAAC;IACjB,OAAO,EAAE,CAAC;CACX,CAAC,CAAA;AAEJ,MAAM,OAAO,UAAW,SAAQ,MAAM,CAAC,KAAK,CAAa,YAAY,CAAC,CAAC;IACrE,IAAI,EAAE,MAAM,CAAC,MAAM;IACnB,EAAE,EAAE,MAAM,CAAC,GAAG;IACd,MAAM,EAAE,WAAW;IACnB,OAAO,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC;IAC1C,QAAQ,EAAE,aAAa;IACvB,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC;CACpD,CAAC;CAAG;AAEL,MAAM,OAAO,UAAW,SAAQ,MAAM,CAAC,KAAK,CAAa,YAAY,CAAC,CAAC;IACrE,IAAI,EAAE,MAAM,CAAC,MAAM;IACnB,KAAK,EAAE,MAAM,CAAC,GAAG;IACjB,GAAG,EAAE,MAAM,CAAC,GAAG;IACf,SAAS,EAAE,MAAM,CAAC,MAAM;CACzB,CAAC;CAAG;AAEL,MAAM,OAAO,UAAW,SAAQ,MAAM,CAAC,KAAK,CAAa,YAAY,CAAC,CAAC;IACrE,QAAQ,EAAE,MAAM,CAAC,MAAM;IACvB,KAAK,EAAE,MAAM,CAAC,MAAM;IACpB,KAAK,EAAE,MAAM,CAAC,OAAO;CACtB,CAAC;CAAG;AAEL,MAAM,OAAO,YAAa,SAAQ,MAAM,CAAC,KAAK,CAAe,cAAc,CAAC,CAAC;IAC3E,QAAQ,EAAE,MAAM,CAAC,MAAM;IACvB,EAAE,EAAE,MAAM,CAAC,MAAM;IACjB,IAAI,EAAE,MAAM,CAAC,MAAM;IACnB,KAAK,EAAE,MAAM,CAAC,GAAG;IACjB,QAAQ,EAAE,MAAM,CAAC,MAAM;IACvB,OAAO,EAAE,MAAM,CAAC,MAAM;CACvB,CAAC;CAAG;AAEL,MAAM,OAAO,YAAa,SAAQ,MAAM,CAAC,KAAK,CAAe,cAAc,CAAC,CAAC;IAC3E,WAAW,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC;IAC/C,QAAQ,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC;IACxC,WAAW,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC;IAC3C,WAAW,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC;IAC9C,YAAY,EAAE,MAAM,CAAC,GAAG;IACxB,SAAS,EAAE,MAAM,CAAC,GAAG;IACrB,iBAAiB,EAAE,MAAM,CAAC,GAAG;IAC7B,SAAS,EAAE,MAAM,CAAC,GAAG;IACrB,QAAQ,EAAE,MAAM,CAAC,GAAG;IACpB,WAAW,EAAE,MAAM,CAAC,GAAG;IACvB,WAAW,EAAE,MAAM,CAAC,GAAG;IACvB,SAAS,EAAE,MAAM,CAAC,GAAG;IACrB,SAAS,EAAE,MAAM,CAAC,GAAG;IACrB,qBAAqB,EAAE,MAAM,CAAC,GAAG;IACjC,aAAa,EAAE,MAAM,CAAC,GAAG;IACzB,WAAW,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC;IAC/C,UAAU,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC;IAC1C,cAAc,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC;IAC9C,WAAW,EAAE,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC;CACtC,CAAC;CAAG;AAEL,MAAM,OAAO,WAAY,SAAQ,MAAM,CAAC,KAAK,CAAc,aAAa,CAAC,CAAC;IACxE,aAAa,EAAE,MAAM,CAAC,GAAG;IACzB,KAAK,EAAE,MAAM,CAAC,MAAM;IACpB,KAAK,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC;IACxC,SAAS,EAAE,MAAM,CAAC,MAAM;IACxB,UAAU,EAAE,MAAM,CAAC,MAAM;IACzB,QAAQ,EAAE,MAAM,CAAC,MAAM;IACvB,cAAc,EAAE,MAAM,CAAC,MAAM;IAC7B,YAAY,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC;IACzC,KAAK,EAAE,UAAU;IACjB,WAAW,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC;IAC9C,aAAa,EAAE,MAAM,CAAC,MAAM;IAC5B,IAAI,EAAE,MAAM,CAAC,MAAM;IACnB,kBAAkB,EAAE,MAAM,CAAC,MAAM;IACjC,OAAO,EAAE,YAAY;IACrB,OAAO,EAAE,MAAM,CAAC,MAAM;IACtB,OAAO,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC;IAC3C,OAAO,EAAE,MAAM,CAAC,GAAG;IACnB,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC;IAChC,OAAO,EAAE,MAAM,CAAC,WAAW,CAAC,YAAY,CAAC;CAC1C,CAAC;IACA,IAAI,SAAS;QACX,OAAO,IAAI,CAAC,OAAO,KAAK,WAAW,CAAA;IACrC,CAAC;IAED,IAAI,WAAW;QACb,OAAO,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,MAAM,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAA;IAC9F,CAAC;IAED,IAAI,YAAY;QACd,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE,CAC3C,KAAK,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CACnD,CAAA;QACD,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,EAAE,CAAC,CAAC,CAAA;IACxF,CAAC;IAED,IAAI,WAAW;QACb,OAAO,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,CAAC,CAAA;IAChF,CAAC;CACF;AAED,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAA;AACnC,MAAM,CAAC,MAAM,aAAa,GAAG,QAAQ,CAAA;AAUrC,MAAM,CAAC,MAAM,qBAAqB,GAAG,GAAqB,EAAE,CAAC,CAAC;IAC5D,UAAU,EAAE,EAAE;IACd,KAAK,EAAE,IAAI,GAAG,EAAE;IAChB,QAAQ,EAAE,IAAI,GAAG,EAAE;IACnB,QAAQ,EAAE,IAAI,GAAG,EAAE;IACnB,YAAY,EAAE,IAAI,GAAG,EAAE;CACxB,CAAC,CAAA;AAEF,MAAM,YAAY,GAAG,CAAC,WAA6B,EAAU,EAAE,CAC7D,WAAW,CAAC,UAAU,CAAC,WAAW,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,aAAa,CAAA;AAE5E,MAAM,SAAS,GAAG,CAAC,MAA6B,EAAE,KAAa,EAAyB,EAAE;IACxF,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAA;IACnC,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAA;AACrF,CAAC,CAAA;AAED,MAAM,cAAc,GAAG,CAAC,OAAe,EAAE,QAAuB,EAA6B,EAAE;IAC7F,IAAI,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE,CAAC;QAClC,OAAO,IAAI,aAAa,CAAC;YACvB,GAAG,QAAQ;YACX,YAAY,EAAE,QAAQ,CAAC,YAAY,GAAG,CAAC;SACxC,CAAC,CAAA;IACJ,CAAC;IACD,IAAI,OAAO,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,CAAC;QACtC,OAAO,IAAI,aAAa,CAAC;YACvB,GAAG,QAAQ;YACX,gBAAgB,EAAE,QAAQ,CAAC,gBAAgB,GAAG,CAAC;SAChD,CAAC,CAAA;IACJ,CAAC;IACD,IAAI,OAAO,CAAC,UAAU,CAAC,eAAe,CAAC,EAAE,CAAC;QACxC,OAAO,IAAI,aAAa,CAAC;YACvB,GAAG,QAAQ;YACX,WAAW,EAAE,QAAQ,CAAC,WAAW,GAAG,CAAC;SACtC,CAAC,CAAA;IACJ,CAAC;IACD,IAAI,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;QACvC,OAAO,IAAI,aAAa,CAAC;YACvB,GAAG,QAAQ;YACX,cAAc,EAAE,QAAQ,CAAC,cAAc,GAAG,CAAC;SAC5C,CAAC,CAAA;IACJ,CAAC;IACD,IAAI,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,EAAE,CAAC;QACrC,OAAO,IAAI,aAAa,CAAC;YACvB,GAAG,QAAQ;YACX,OAAO,EAAE,QAAQ,CAAC,OAAO,GAAG,CAAC;SAC9B,CAAC,CAAA;IACJ,CAAC;IACD,OAAO,SAAS,CAAA;AAClB,CAAC,CAAA;AAED,MAAM,CAAC,MAAM,YAAY,GAAG,CAAC,WAA6B,EAAE,KAAgB,EAAoB,EAAE;IAChG,QAAQ,KAAK,CAAC,IAAI,EAAE,CAAC;QACnB,KAAK,cAAc;YACjB,OAAO;gBACL,GAAG,WAAW;gBACd,UAAU,EAAE,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,WAAW,CAAC,UAAU,CAAC;aACrD,CAAA;QACH,KAAK,gBAAgB,CAAC;QACtB,KAAK,aAAa;YAChB,OAAO;gBACL,GAAG,WAAW;gBACd,UAAU,EAAE,SAAS,CAAC,WAAW,CAAC,UAAU,EAAE,KAAK,CAAC,KAAK,CAAC;aAC3D,CAAA;QACH,KAAK,YAAY,CAAC,CAAC,CAAC;YAClB,MAAM,KAAK,GAAG,YAAY,CAAC,WAAW,CAAC,CAAA;YACvC,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,IAAI,EAAE,CAAA;YAC/B,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAA;YAC1C,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,WAAW,CAAC,KAAK,CAAC,CAAA;YACxC,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,gBAAgB,EAAE,CAAC,CAAC,GAAG,CAAC,oBAAoB,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAA;YAC7F,OAAO;gBACL,GAAG,WAAW;gBACd,KAAK;gBACL,QAAQ,EAAE,IAAI,GAAG,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;gBAChE,YAAY,EACV,KAAK,CAAC,KAAK,KAAK,SAAS;oBACvB,CAAC,CAAC,WAAW,CAAC,YAAY;oBAC1B,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,GAAG,WAAW,CAAC,YAAY,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;aAC1D,CAAA;QACH,CAAC;QACD,KAAK,MAAM,CAAC,CAAC,CAAC;YACZ,MAAM,KAAK,GAAG,YAAY,CAAC,WAAW,CAAC,CAAA;YACvC,MAAM,OAAO,GAAG,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,kBAAkB,EAAE,CAAA;YACvE,MAAM,IAAI,GAAG,cAAc,CAAC,KAAK,CAAC,OAAO,EAAE,OAAO,CAAC,CAAA;YACnD,OAAO,IAAI,KAAK,SAAS;gBACvB,CAAC,CAAC,WAAW;gBACb,CAAC,CAAC;oBACE,GAAG,WAAW;oBACd,QAAQ,EAAE,IAAI,GAAG,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,KAAK,EAAE,IAAI,CAAC;iBACzD,CAAA;QACP,CAAC;QACD;YACE,OAAO,WAAW,CAAA;IACtB,CAAC;AACH,CAAC,CAAA;AAED,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,WAA6B,EAAE,KAAa,EAAe,EAAE,CAC5F,CAAC,GAAG,WAAW,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,CAAC,GAAG,EAAE,MAAM,CAAC,EAAE,EAAE;IAC/D,MAAM,KAAK,GAAG,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;IAC3C,OAAO,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAA;AACzD,CAAC,EAAE,gBAAgB,EAAE,CAAC,CAAA;AAExB,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,WAA6B,EAAE,KAAa,EAAiB,EAAE,CAChG,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,kBAAkB,EAAE,CAAA;AAEzD,MAAM,CAAC,MAAM,cAAc,GAAG,CAC5B,WAA6B,EAC7B,KAAa,EACb,EAAU,EACV,SAA2C,EAAE,EACjC,EAAE;IACd,MAAM,KAAK,GAAG,CAAC,GAAG,WAAW,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,EAAE,MAAM,CAAC,EAAE,EAAE;QACvE,MAAM,KAAK,GAAG,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;QAC3C,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,EAAE,CAAC;YACzB,OAAO,EAAE,CAAA;QACX,CAAC;QACD,MAAM,IAAI,GAAG,eAAe,CAC1B,KAAK,CAAC,CAAC,CAAC,EACR,UAAU,CAAC,IAAI,CAAC;YACd,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,UAAU,EAAE,MAAM,CAAC,UAAU;YAC7B,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,GAAG,CAAC,MAAM,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC;SACzD,CAAC,CACH,CAAA;QACD,OAAO,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAA;IACzC,CAAC,CAAC,CAAA;IACF,OAAO,UAAU,CAAC,IAAI,CAAC;QACrB,IAAI,EAAE,KAAK;QACX,EAAE;QACF,MAAM,EAAE,gBAAgB,CAAC,WAAW,EAAE,KAAK,CAAC;QAC5C,QAAQ,EAAE,kBAAkB,CAAC,WAAW,EAAE,KAAK,CAAC;QAChD,MAAM;QACN,GAAG,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC;KACvF,CAAC,CAAA;AACJ,CAAC,CAAA;AAED,MAAM,KAAK,GAAG,CAAC,MAAuB,EAA0B,EAAE,CAChE,WAAW,IAAI,MAAM,IAAI,gBAAgB,IAAI,MAAM,IAAI,QAAQ,IAAI,MAAM,CAAA;AAO3E,MAAM,CAAC,MAAM,YAAY,GAAG,MAAM,CAAC,EAAE,CAAC,4BAA4B,CAAC,CAAC,QAAQ,CAAC,EAC3E,MAAuB;IAEvB,MAAM,WAAW,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,qBAAqB,EAAE,CAAC,CAAA;IAC5D,MAAM,QAAQ,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;IACnC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC;QACnB,OAAO;YACL,GAAG,EAAE,GAAG,CAAC,GAAG,CAAC,WAAW,CAAC;YACzB,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,WAAW,EAAE,qBAAqB,EAAE,CAAC;SACrD,CAAA;IACH,CAAC;IACD,MAAM,YAAY,GAAG,KAAK,CAAC,CAAC,MAAM,CAAC,SAAS,CAAA;IAC5C,MAAM,QAAQ,GAAG,KAAK,CAAC,CAAC,MAAM,CAAC,cAAc,CAAA;IAC7C,KAAK,CAAC,CAAC,MAAM,CAAC,gBAAgB,CAAC,YAAY,CAAC,CAAC,IAAI,CAC/C,MAAM,CAAC,UAAU,CAAC,CAAC,KAAK,EAAE,EAAE,CAC1B,GAAG,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,YAAY,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,IAAI,CACrE,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAC3D,CACF,EACD,MAAM,CAAC,UAAU,CAClB,CAAA;IACD,MAAM,KAAK,GAAwB,MAAM,CAAC,OAAO,CAAC,GAAG,EAAE,CACrD,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,cAAc,EAAE,GAAG,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CACzD,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,EAAE,IAAI,CAAC,EAAE,EAAE,CACnC,IAAI,IAAI,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CACzF,CACF,CACF,CAAA;IACD,OAAO;QACL,GAAG,EAAE,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC;QACrD,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,WAAW,EAAE,qBAAqB,EAAE,CAAC,CAAC,CAAC;KACjF,CAAA;AACH,CAAC,CAAC,CAAA"}
|
|
@@ -0,0 +1,9 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import { BenchRecord } from "./Bench.ts";
|
|
3
|
+
import { PlanParseError, UnsupportedSchemaVersion } from "./FlowError.ts";
|
|
4
|
+
import type { PlainFileStoreShape } from "./Persistence.ts";
|
|
5
|
+
export declare const median: (values: ReadonlyArray<number>) => number;
|
|
6
|
+
export declare const renderBenchReport: (records: ReadonlyArray<BenchRecord>, projectPrograms?: number) => string;
|
|
7
|
+
export declare const appendBenchRecord: (files: PlainFileStoreShape, path: string, record: BenchRecord) => Effect.Effect<void, import("./FlowError.ts").PersistenceError | PlanParseError | UnsupportedSchemaVersion | import("./FlowError.ts").WorkspacePathError | import("./FlowError.ts").WorkspaceLimitError | import("./FlowError.ts").WorkspaceIoError | import("./FlowError.ts").FlowAborted | import("./FlowError.ts").ProcessError | import("./FlowError.ts").FlowLlmError | import("./FlowError.ts").FlowCapabilityDenied | import("./FlowError.ts").BudgetExceeded, never>;
|
|
8
|
+
export declare const loadBenchRecords: (files: PlainFileStoreShape, path: string) => Effect.Effect<readonly BenchRecord[], import("./FlowError.ts").PersistenceError | PlanParseError | UnsupportedSchemaVersion | import("./FlowError.ts").WorkspacePathError | import("./FlowError.ts").WorkspaceLimitError | import("./FlowError.ts").WorkspaceIoError | import("./FlowError.ts").FlowAborted | import("./FlowError.ts").ProcessError | import("./FlowError.ts").FlowLlmError | import("./FlowError.ts").FlowCapabilityDenied | import("./FlowError.ts").BudgetExceeded, never>;
|
|
9
|
+
//# sourceMappingURL=BenchReport.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"BenchReport.d.ts","sourceRoot":"","sources":["../src/BenchReport.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAEvC,OAAO,EAAE,WAAW,EAAuC,MAAM,YAAY,CAAA;AAC7E,OAAO,EAAE,cAAc,EAAE,wBAAwB,EAAkB,MAAM,gBAAgB,CAAA;AACzF,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAA;AAI3D,eAAO,MAAM,MAAM,GAAI,QAAQ,aAAa,CAAC,MAAM,CAAC,KAAG,MAStD,CAAA;AA4WD,eAAO,MAAM,iBAAiB,GAC5B,SAAS,aAAa,CAAC,WAAW,CAAC,EACnC,kBAAkB,MAAM,KACvB,MAUF,CAAA;AAID,eAAO,MAAM,iBAAiB,ghBAa5B,CAAA;AAEF,eAAO,MAAM,gBAAgB,6gBAiC3B,CAAA"}
|
|
@@ -0,0 +1,360 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import * as Schema from "effect/Schema";
|
|
3
|
+
import { BenchRecord, CurrentBenchSchema } from "./Bench.js";
|
|
4
|
+
import { PlanParseError, UnsupportedSchemaVersion } from "./FlowError.js";
|
|
5
|
+
import { estimateCostUsd } from "./PriceList.js";
|
|
6
|
+
import { TokenUsage } from "@llm4ts/core/Models";
|
|
7
|
+
export const median = (values) => {
|
|
8
|
+
if (values.length === 0) {
|
|
9
|
+
return 0;
|
|
10
|
+
}
|
|
11
|
+
const sorted = [...values].sort((left, right) => left - right);
|
|
12
|
+
const middle = Math.floor(sorted.length / 2);
|
|
13
|
+
return sorted.length % 2 === 1
|
|
14
|
+
? (sorted[middle] ?? 0)
|
|
15
|
+
: ((sorted[middle - 1] ?? 0) + (sorted[middle] ?? 0)) / 2;
|
|
16
|
+
};
|
|
17
|
+
const formatDuration = (milliseconds) => {
|
|
18
|
+
if (milliseconds >= 3_600_000) {
|
|
19
|
+
return `${(milliseconds / 3_600_000).toFixed(1)}h`;
|
|
20
|
+
}
|
|
21
|
+
if (milliseconds >= 60_000) {
|
|
22
|
+
return `${(milliseconds / 60_000).toFixed(1)}m`;
|
|
23
|
+
}
|
|
24
|
+
return `${(milliseconds / 1_000).toFixed(1)}s`;
|
|
25
|
+
};
|
|
26
|
+
const formatTokens = (tokens) => {
|
|
27
|
+
if (Math.abs(tokens) >= 1_000_000) {
|
|
28
|
+
return `${(tokens / 1_000_000).toFixed(1)}M`;
|
|
29
|
+
}
|
|
30
|
+
if (Math.abs(tokens) >= 1_000) {
|
|
31
|
+
return `${(tokens / 1_000).toFixed(1)}k`;
|
|
32
|
+
}
|
|
33
|
+
return Math.round(tokens).toString();
|
|
34
|
+
};
|
|
35
|
+
const formatInteger = (value) => Math.round(value).toString();
|
|
36
|
+
const formatPercent = (value) => `${value.toFixed(1)}%`;
|
|
37
|
+
const formatBoolean = (value) => (value >= 0.5 ? "yes" : "no");
|
|
38
|
+
const spread = (records, metric) => {
|
|
39
|
+
const values = records.flatMap((record) => {
|
|
40
|
+
const value = metric.value(record);
|
|
41
|
+
return value === undefined ? [] : [value];
|
|
42
|
+
});
|
|
43
|
+
if (values.length === 0) {
|
|
44
|
+
return "–";
|
|
45
|
+
}
|
|
46
|
+
const center = median(values);
|
|
47
|
+
const base = metric.format(center);
|
|
48
|
+
return values.length > 1 && Math.min(...values) !== Math.max(...values)
|
|
49
|
+
? `${base} (${metric.format(Math.min(...values))}–${metric.format(Math.max(...values))})`
|
|
50
|
+
: base;
|
|
51
|
+
};
|
|
52
|
+
const judgePercent = (record) => {
|
|
53
|
+
const scores = record.quality?.judgeScores ?? [];
|
|
54
|
+
const maximum = scores.reduce((sum, score) => sum + score.max, 0);
|
|
55
|
+
return maximum <= 0
|
|
56
|
+
? undefined
|
|
57
|
+
: (scores.reduce((sum, score) => sum + score.score, 0) / maximum) * 100;
|
|
58
|
+
};
|
|
59
|
+
const totalOutputTokens = (record) => record.phases.reduce((sum, phase) => sum + phase.tokens.completion, 0);
|
|
60
|
+
const totalCachedTokens = (record) => record.phases.reduce((sum, phase) => sum + phase.tokens.cached, 0);
|
|
61
|
+
const headlineMetrics = (judgesComparable) => [
|
|
62
|
+
{
|
|
63
|
+
label: "Total time",
|
|
64
|
+
direction: "lower",
|
|
65
|
+
value: (record) => record.totalMs,
|
|
66
|
+
format: formatDuration
|
|
67
|
+
},
|
|
68
|
+
{
|
|
69
|
+
label: "Total tokens",
|
|
70
|
+
direction: "lower",
|
|
71
|
+
value: (record) => record.totalTokens,
|
|
72
|
+
format: formatTokens
|
|
73
|
+
},
|
|
74
|
+
{
|
|
75
|
+
label: "Output tokens",
|
|
76
|
+
direction: "lower",
|
|
77
|
+
value: totalOutputTokens,
|
|
78
|
+
format: formatTokens
|
|
79
|
+
},
|
|
80
|
+
{
|
|
81
|
+
label: "Cached reads",
|
|
82
|
+
direction: "neutral",
|
|
83
|
+
value: totalCachedTokens,
|
|
84
|
+
format: formatTokens
|
|
85
|
+
},
|
|
86
|
+
{
|
|
87
|
+
label: "Self-healing actions",
|
|
88
|
+
direction: "lower",
|
|
89
|
+
value: (record) => record.selfHealing,
|
|
90
|
+
format: formatInteger
|
|
91
|
+
},
|
|
92
|
+
{
|
|
93
|
+
label: "Gate cleared",
|
|
94
|
+
direction: "higher",
|
|
95
|
+
value: (record) => record.quality?.gateCleared === undefined ? undefined : record.quality.gateCleared ? 1 : 0,
|
|
96
|
+
format: formatBoolean,
|
|
97
|
+
gap: formatInteger
|
|
98
|
+
},
|
|
99
|
+
{
|
|
100
|
+
label: "Build passed",
|
|
101
|
+
direction: "higher",
|
|
102
|
+
value: (record) => record.quality?.buildPassed === undefined ? undefined : record.quality.buildPassed ? 1 : 0,
|
|
103
|
+
format: formatBoolean,
|
|
104
|
+
gap: formatInteger
|
|
105
|
+
},
|
|
106
|
+
{
|
|
107
|
+
label: "Coverage %",
|
|
108
|
+
direction: "higher",
|
|
109
|
+
value: (record) => record.quality?.coveragePct,
|
|
110
|
+
format: formatPercent
|
|
111
|
+
},
|
|
112
|
+
{
|
|
113
|
+
label: "Judge score",
|
|
114
|
+
direction: judgesComparable ? "higher" : "neutral",
|
|
115
|
+
value: judgePercent,
|
|
116
|
+
format: formatPercent
|
|
117
|
+
}
|
|
118
|
+
];
|
|
119
|
+
const gapCell = (metric, values) => {
|
|
120
|
+
if (values.length < 2) {
|
|
121
|
+
return "";
|
|
122
|
+
}
|
|
123
|
+
const low = Math.min(...values);
|
|
124
|
+
const high = Math.max(...values);
|
|
125
|
+
const difference = high - low;
|
|
126
|
+
const percent = low === 0 ? 100 : (difference / Math.abs(low)) * 100;
|
|
127
|
+
return `+${(metric.gap ?? metric.format)(difference)} (+${percent.toFixed(1)}%)`;
|
|
128
|
+
};
|
|
129
|
+
const metricRow = (metric, providers, byProvider) => {
|
|
130
|
+
const centers = providers.map((provider) => {
|
|
131
|
+
const values = (byProvider.get(provider) ?? []).flatMap((record) => {
|
|
132
|
+
const value = metric.value(record);
|
|
133
|
+
return value === undefined ? [] : [value];
|
|
134
|
+
});
|
|
135
|
+
return values.length === 0 ? undefined : median(values);
|
|
136
|
+
});
|
|
137
|
+
const comparable = centers.filter((value) => value !== undefined);
|
|
138
|
+
const best = metric.direction === "neutral" || comparable.length < 2
|
|
139
|
+
? undefined
|
|
140
|
+
: metric.direction === "lower"
|
|
141
|
+
? Math.min(...comparable)
|
|
142
|
+
: Math.max(...comparable);
|
|
143
|
+
const worst = metric.direction === "neutral" || comparable.length < 2
|
|
144
|
+
? undefined
|
|
145
|
+
: metric.direction === "lower"
|
|
146
|
+
? Math.max(...comparable)
|
|
147
|
+
: Math.min(...comparable);
|
|
148
|
+
const cells = providers.map((provider, index) => {
|
|
149
|
+
const value = centers[index];
|
|
150
|
+
const display = spread(byProvider.get(provider) ?? [], metric);
|
|
151
|
+
if (value === undefined || best === undefined || best === worst) {
|
|
152
|
+
return display;
|
|
153
|
+
}
|
|
154
|
+
return value === best ? `**${display} ✅**` : value === worst ? `${display} ⚠️` : display;
|
|
155
|
+
});
|
|
156
|
+
return `| ${metric.label} | ${cells.join(" | ")} | ${gapCell(metric, comparable)} |`;
|
|
157
|
+
};
|
|
158
|
+
const costForPhase = (record, phase) => {
|
|
159
|
+
if (phase.costUsd !== undefined) {
|
|
160
|
+
return [phase.costUsd, false];
|
|
161
|
+
}
|
|
162
|
+
if (phase.tokens.prompt === 0 && phase.tokens.completion === 0 && phase.tokens.cached === 0) {
|
|
163
|
+
return undefined;
|
|
164
|
+
}
|
|
165
|
+
const cost = estimateCostUsd(record.modelRequested, TokenUsage.make({
|
|
166
|
+
prompt: phase.tokens.prompt,
|
|
167
|
+
completion: phase.tokens.completion,
|
|
168
|
+
total: phase.tokens.total,
|
|
169
|
+
...(phase.tokens.cached <= 0 ? {} : { cached: phase.tokens.cached })
|
|
170
|
+
}));
|
|
171
|
+
return cost === undefined ? undefined : [cost, true];
|
|
172
|
+
};
|
|
173
|
+
const costForRecord = (record) => {
|
|
174
|
+
const values = record.phases.flatMap((phase) => {
|
|
175
|
+
const cost = costForPhase(record, phase);
|
|
176
|
+
return cost === undefined ? [] : [cost];
|
|
177
|
+
});
|
|
178
|
+
return values.length === 0
|
|
179
|
+
? undefined
|
|
180
|
+
: [values.reduce((sum, [cost]) => sum + cost, 0), values.some(([, estimated]) => estimated)];
|
|
181
|
+
};
|
|
182
|
+
const costRow = (providers, byProvider, select) => {
|
|
183
|
+
const cells = providers.map((provider) => {
|
|
184
|
+
const values = (byProvider.get(provider) ?? []).flatMap((record) => {
|
|
185
|
+
const value = select(record);
|
|
186
|
+
return value === undefined ? [] : [value];
|
|
187
|
+
});
|
|
188
|
+
if (values.length === 0) {
|
|
189
|
+
return "–";
|
|
190
|
+
}
|
|
191
|
+
const suffix = values.some(([, estimated]) => estimated) ? "*" : "";
|
|
192
|
+
const costs = values.map(([cost]) => cost);
|
|
193
|
+
return `$${median(costs).toFixed(2)}${suffix}`;
|
|
194
|
+
});
|
|
195
|
+
return `| Est. cost (USD) | ${cells.join(" | ")} | |`;
|
|
196
|
+
};
|
|
197
|
+
const projection = (programs, providers, byProvider) => {
|
|
198
|
+
const rows = providers.map((provider) => {
|
|
199
|
+
const records = byProvider.get(provider) ?? [];
|
|
200
|
+
const perProgramTokens = records.flatMap((record) => {
|
|
201
|
+
const count = record.quality?.programs ?? 0;
|
|
202
|
+
return count <= 0 ? [] : [record.totalTokens / count];
|
|
203
|
+
});
|
|
204
|
+
const perProgramMs = records.flatMap((record) => {
|
|
205
|
+
const count = record.quality?.programs ?? 0;
|
|
206
|
+
return count <= 0 ? [] : [record.totalMs / count];
|
|
207
|
+
});
|
|
208
|
+
if (perProgramTokens.length === 0 || perProgramMs.length === 0) {
|
|
209
|
+
return `| ${provider} | – | – | – |`;
|
|
210
|
+
}
|
|
211
|
+
const tokens = median(perProgramTokens);
|
|
212
|
+
const milliseconds = median(perProgramMs);
|
|
213
|
+
return `| ${provider} | ${formatTokens(tokens)} | ${formatTokens(tokens * programs)} | ${formatDuration(milliseconds * programs)} |`;
|
|
214
|
+
});
|
|
215
|
+
return [
|
|
216
|
+
`## Projection @ ${programs} programs`,
|
|
217
|
+
"",
|
|
218
|
+
"| Provider | Tokens/program | Projected tokens | Projected time |",
|
|
219
|
+
"| -------- | -------------- | ---------------- | -------------- |",
|
|
220
|
+
...rows,
|
|
221
|
+
"",
|
|
222
|
+
"_Linear extrapolation from recorded medians; not a delivery forecast._"
|
|
223
|
+
].join("\n");
|
|
224
|
+
};
|
|
225
|
+
const renderSection = (records, projectPrograms) => {
|
|
226
|
+
const providers = [...new Set(records.map((record) => record.provider))].sort();
|
|
227
|
+
const byProvider = new Map(providers.map((provider) => [
|
|
228
|
+
provider,
|
|
229
|
+
records.filter((record) => record.provider === provider)
|
|
230
|
+
]));
|
|
231
|
+
const first = records[0];
|
|
232
|
+
if (first === undefined) {
|
|
233
|
+
return "";
|
|
234
|
+
}
|
|
235
|
+
const judgesComparable = new Set(records.map((record) => `${record.judge.provider}\0${record.judge.model}`)).size === 1;
|
|
236
|
+
const judgeLine = judgesComparable
|
|
237
|
+
? `Judge: ${first.judge.provider}/${first.judge.model} — same examiner for all runs`
|
|
238
|
+
: "Judges: self-graded — scores NOT cross-comparable";
|
|
239
|
+
const header = `| Metric | ${providers.join(" | ")} | Gap |`;
|
|
240
|
+
const separator = `| ------ | ${providers.map(() => "------").join(" | ")} | --- |`;
|
|
241
|
+
const outcomes = `| Outcome | ${providers
|
|
242
|
+
.map((provider) => {
|
|
243
|
+
const group = byProvider.get(provider) ?? [];
|
|
244
|
+
return group.some((record) => record.resumed === true)
|
|
245
|
+
? `${group[0]?.outcome ?? "unknown"} ♻`
|
|
246
|
+
: (group[0]?.outcome ?? "unknown");
|
|
247
|
+
})
|
|
248
|
+
.join(" | ")} | |`;
|
|
249
|
+
const phaseNames = [
|
|
250
|
+
...new Set(records.flatMap((record) => record.phases.map((phase) => phase.name)))
|
|
251
|
+
];
|
|
252
|
+
const phaseSections = phaseNames.map((phaseName) => {
|
|
253
|
+
const duration = {
|
|
254
|
+
label: "Duration",
|
|
255
|
+
direction: "lower",
|
|
256
|
+
value: (record) => record.phases.find((phase) => phase.name === phaseName)?.ms,
|
|
257
|
+
format: formatDuration
|
|
258
|
+
};
|
|
259
|
+
const tokens = {
|
|
260
|
+
label: "Tokens",
|
|
261
|
+
direction: "lower",
|
|
262
|
+
value: (record) => {
|
|
263
|
+
const phase = record.phases.find((item) => item.name === phaseName);
|
|
264
|
+
return phase === undefined ? undefined : phase.tokens.total + phase.tokens.cached;
|
|
265
|
+
},
|
|
266
|
+
format: formatTokens
|
|
267
|
+
};
|
|
268
|
+
return [
|
|
269
|
+
`### ${phaseName}`,
|
|
270
|
+
"",
|
|
271
|
+
header,
|
|
272
|
+
separator,
|
|
273
|
+
metricRow(duration, providers, byProvider),
|
|
274
|
+
metricRow(tokens, providers, byProvider),
|
|
275
|
+
costRow(providers, byProvider, (record) => {
|
|
276
|
+
const phase = record.phases.find((item) => item.name === phaseName);
|
|
277
|
+
return phase === undefined ? undefined : costForPhase(record, phase);
|
|
278
|
+
})
|
|
279
|
+
].join("\n");
|
|
280
|
+
});
|
|
281
|
+
const failures = records.filter((record) => !record.completed);
|
|
282
|
+
const estimated = records.some((record) => costForRecord(record)?.[1] === true);
|
|
283
|
+
const extras = [
|
|
284
|
+
...(failures.length === 0
|
|
285
|
+
? []
|
|
286
|
+
: [
|
|
287
|
+
"## Failures",
|
|
288
|
+
...failures.map((record) => `- ${record.provider} ${record.runId}: ${record.outcome}`)
|
|
289
|
+
]),
|
|
290
|
+
...(projectPrograms === undefined ? [] : [projection(projectPrograms, providers, byProvider)]),
|
|
291
|
+
...(records.some((record) => record.resumed === true)
|
|
292
|
+
? [
|
|
293
|
+
"_♻ resumed run: duration and tokens include only work after resume and are not fresh-run comparable._"
|
|
294
|
+
]
|
|
295
|
+
: []),
|
|
296
|
+
...(estimated
|
|
297
|
+
? ["_\\* Cost estimated at report time from recorded tokens × the requested model._"]
|
|
298
|
+
: []),
|
|
299
|
+
`_Machines: ${[...new Set(records.map((record) => record.machine.hostname))].sort().join(", ")}._`
|
|
300
|
+
];
|
|
301
|
+
return [
|
|
302
|
+
`# Benchmark — ${first.pack} @ ${first.fixtureFingerprint.slice(0, 8)}`,
|
|
303
|
+
"",
|
|
304
|
+
`_${judgeLine}_`,
|
|
305
|
+
"",
|
|
306
|
+
"## Headline",
|
|
307
|
+
"",
|
|
308
|
+
header,
|
|
309
|
+
separator,
|
|
310
|
+
outcomes,
|
|
311
|
+
...headlineMetrics(judgesComparable).map((metric) => metricRow(metric, providers, byProvider)),
|
|
312
|
+
costRow(providers, byProvider, costForRecord),
|
|
313
|
+
"",
|
|
314
|
+
"## Phases",
|
|
315
|
+
"",
|
|
316
|
+
...phaseSections,
|
|
317
|
+
...extras
|
|
318
|
+
].join("\n");
|
|
319
|
+
};
|
|
320
|
+
export const renderBenchReport = (records, projectPrograms) => {
|
|
321
|
+
const groups = new Map();
|
|
322
|
+
for (const record of records) {
|
|
323
|
+
const key = `${record.pack}\0${record.fixtureFingerprint}`;
|
|
324
|
+
groups.set(key, [...(groups.get(key) ?? []), record]);
|
|
325
|
+
}
|
|
326
|
+
return [...groups.entries()]
|
|
327
|
+
.sort(([left], [right]) => left.localeCompare(right))
|
|
328
|
+
.map(([, group]) => renderSection(group, projectPrograms))
|
|
329
|
+
.join("\n\n");
|
|
330
|
+
};
|
|
331
|
+
const benchCodec = Schema.fromJsonString(BenchRecord);
|
|
332
|
+
export const appendBenchRecord = Effect.fn("@llm4ts/flow/BenchReport.append")(function* (files, path, record) {
|
|
333
|
+
const encoded = yield* Schema.encodeEffect(benchCodec)(record).pipe(Effect.mapError((error) => PlanParseError.make({
|
|
334
|
+
message: `failed to encode benchmark record: ${String(error)}`
|
|
335
|
+
})));
|
|
336
|
+
yield* files.append(path, `${encoded}\n`);
|
|
337
|
+
});
|
|
338
|
+
export const loadBenchRecords = Effect.fn("@llm4ts/flow/BenchReport.load")(function* (files, path) {
|
|
339
|
+
const contents = yield* files.read(path);
|
|
340
|
+
if (contents === undefined) {
|
|
341
|
+
return [];
|
|
342
|
+
}
|
|
343
|
+
const records = yield* Effect.forEach(contents
|
|
344
|
+
.split(/\r?\n/u)
|
|
345
|
+
.map((line) => line.trim())
|
|
346
|
+
.filter((line) => line.length > 0), (line, index) => Schema.decodeUnknownEffect(benchCodec)(line).pipe(Effect.mapError((error) => PlanParseError.make({
|
|
347
|
+
message: `malformed benchmark record at ${path}:${index + 1}: ${String(error)}`
|
|
348
|
+
}))), { concurrency: 1 });
|
|
349
|
+
for (const record of records) {
|
|
350
|
+
if (record.schemaVersion !== CurrentBenchSchema) {
|
|
351
|
+
return yield* UnsupportedSchemaVersion.make({
|
|
352
|
+
path,
|
|
353
|
+
expected: CurrentBenchSchema,
|
|
354
|
+
actual: record.schemaVersion
|
|
355
|
+
});
|
|
356
|
+
}
|
|
357
|
+
}
|
|
358
|
+
return records;
|
|
359
|
+
});
|
|
360
|
+
//# sourceMappingURL=BenchReport.js.map
|