adaptive-memory-multi-model-router 2.2.9 → 2.4.0

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.
Files changed (43) hide show
  1. package/README.md +81 -902
  2. package/package.json +1 -1
  3. package/src/skills/__tests__/skill_manager.test.ts +328 -0
  4. package/assets/benchmark-results.png +0 -0
  5. package/assets/complexity-scoring-v2.png +0 -0
  6. package/assets/complexity-scoring.png +0 -0
  7. package/assets/cost-comparison-chart.png +0 -0
  8. package/assets/cost-comparison-v2.png +0 -0
  9. package/assets/feature-comparison-v2.png +0 -0
  10. package/assets/feature-comparison-v3.png +0 -0
  11. package/assets/provider-health-chart.png +0 -0
  12. package/assets/provider-health-v2.png +0 -0
  13. package/assets/routing-flow-v2.png +0 -0
  14. package/assets/routing-flow-v3.png +0 -0
  15. package/assets/routing-flow.png +0 -0
  16. package/assets/tier-distribution.png +0 -0
  17. package/dist/cache/cacheKeyGenerator.d.ts +0 -67
  18. package/dist/cache/cacheKeyGenerator.d.ts.map +0 -1
  19. package/dist/cache/cacheKeyGenerator.js +0 -211
  20. package/dist/cache/cacheKeyGenerator.js.map +0 -1
  21. package/dist/cost/preCallCostEstimator.d.ts +0 -114
  22. package/dist/cost/preCallCostEstimator.d.ts.map +0 -1
  23. package/dist/cost/preCallCostEstimator.js +0 -256
  24. package/dist/cost/preCallCostEstimator.js.map +0 -1
  25. package/dist/inference/speculativeDecoding.d.ts +0 -133
  26. package/dist/inference/speculativeDecoding.d.ts.map +0 -1
  27. package/dist/inference/speculativeDecoding.js +0 -276
  28. package/dist/inference/speculativeDecoding.js.map +0 -1
  29. package/dist/providers/providerHealth.d.ts +0 -117
  30. package/dist/providers/providerHealth.d.ts.map +0 -1
  31. package/dist/providers/providerHealth.js +0 -309
  32. package/dist/providers/providerHealth.js.map +0 -1
  33. package/dist/routing/difficultyClassifier.d.ts +0 -79
  34. package/dist/routing/difficultyClassifier.d.ts.map +0 -1
  35. package/dist/routing/difficultyClassifier.js +0 -329
  36. package/dist/routing/difficultyClassifier.js.map +0 -1
  37. package/dist/sdk.d.ts +0 -125
  38. package/docs/HN_CAMPAIGN.md +0 -785
  39. package/src/cache/cacheKeyGenerator.ts +0 -242
  40. package/src/cost/preCallCostEstimator.ts +0 -345
  41. package/src/inference/speculativeDecoding.ts +0 -373
  42. package/src/providers/providerHealth.ts +0 -397
  43. package/src/routing/difficultyClassifier.ts +0 -420
@@ -1,276 +0,0 @@
1
- "use strict";
2
- /**
3
- * A3M Router - Speculative Decoding Interface
4
- *
5
- * Interface for integrating speculative decoding.
6
- * Currently a stub/interface for future Medusa/Lookahead integration.
7
- *
8
- * Speculative decoding uses a smaller "draft" model to predict
9
- * multiple tokens ahead, which are then verified in parallel by
10
- * the main model. This can provide 2-3x speedup in generation.
11
- *
12
- * Usage:
13
- * const specDec = new SpeculativeDecoding();
14
- * if (specDec.shouldUse(true)) {
15
- * const draftModel = specDec.getDraftModel('medusa');
16
- * // ... use draft model to generate and verify
17
- * }
18
- */
19
- Object.defineProperty(exports, "__esModule", { value: true });
20
- exports.SpeculativeDecodingWrapper = exports.SpeculativeDecoding = void 0;
21
- exports.createSpeculativeDecoding = createSpeculativeDecoding;
22
- // ============================================================
23
- // Speculative Decoding Interface
24
- // ============================================================
25
- class SpeculativeDecoding {
26
- config;
27
- draftModels = new Map();
28
- constructor(config) {
29
- this.config = {
30
- enabled: config?.enabled ?? false,
31
- draftModelType: config?.draftModelType ?? 'medusa',
32
- speculationWindow: config?.speculationWindow ?? 4,
33
- temperature: config?.temperature ?? 0.0,
34
- provider: config?.provider,
35
- model: config?.model,
36
- };
37
- // Register available draft models
38
- this.registerDraftModels();
39
- }
40
- /**
41
- * Determine if speculative decoding should be used.
42
- * Checks config, provider support, and model availability.
43
- */
44
- shouldUse(forceEnable) {
45
- if (forceEnable)
46
- return true;
47
- if (!this.config.enabled)
48
- return false;
49
- // Speculative decoding beneficial for longer outputs
50
- // Check if provider supports it
51
- return this.config.enabled;
52
- }
53
- /**
54
- * Get the draft model configuration.
55
- */
56
- getDraftModel(type) {
57
- const modelType = type || this.config.draftModelType || 'medusa';
58
- const model = this.draftModels.get(modelType);
59
- if (!model) {
60
- // Return a generic configuration
61
- return {
62
- id: `draft-${modelType}`,
63
- type: modelType,
64
- provider: this.config.provider || 'auto',
65
- size: 'small',
66
- supportedWindows: [2, 4, 6, 8],
67
- };
68
- }
69
- return model;
70
- }
71
- /**
72
- * Generate draft tokens using the draft model.
73
- * This is an async method to support API-based draft models.
74
- */
75
- async generateDraft(prompt, maxTokens) {
76
- const draftModel = this.getDraftModel();
77
- if (!draftModel) {
78
- throw new Error('No draft model available');
79
- }
80
- // Stub: In production, this would call the draft model API
81
- // For now, return empty draft
82
- return {
83
- tokens: [],
84
- scores: [],
85
- };
86
- }
87
- /**
88
- * Verify draft tokens against the main model.
89
- * Returns which tokens were accepted and corrections if needed.
90
- */
91
- async verifyDraft(prompt, draftTokens) {
92
- if (draftTokens.length === 0) {
93
- return {
94
- allAccepted: true,
95
- acceptedIndices: [],
96
- rejectedIndices: [],
97
- actualTokens: [],
98
- rewindCount: 0,
99
- };
100
- }
101
- // Stub: In production, this would verify each draft token
102
- // For now, accept all tokens (optimistic)
103
- return {
104
- allAccepted: true,
105
- acceptedIndices: draftTokens.map((_, i) => i),
106
- rejectedIndices: [],
107
- actualTokens: draftTokens,
108
- rewindCount: 0,
109
- };
110
- }
111
- /**
112
- * Calculate speedup from speculative decoding results.
113
- */
114
- calculateSpeedup(result) {
115
- if (!result.used || result.draftTokens === 0)
116
- return 1.0;
117
- // Approximate speedup based on acceptance rate
118
- // Higher acceptance = more speedup
119
- const acceptanceRate = result.acceptanceRate;
120
- // Theoretical 2-3x speedup at 90%+ acceptance
121
- if (acceptanceRate >= 0.9)
122
- return 2.5;
123
- if (acceptanceRate >= 0.8)
124
- return 2.0;
125
- if (acceptanceRate >= 0.7)
126
- return 1.7;
127
- if (acceptanceRate >= 0.5)
128
- return 1.4;
129
- if (acceptanceRate >= 0.3)
130
- return 1.2;
131
- return 1.1;
132
- }
133
- /**
134
- * Update configuration.
135
- */
136
- updateConfig(updates) {
137
- this.config = { ...this.config, ...updates };
138
- }
139
- /**
140
- * Get current configuration.
141
- */
142
- getConfig() {
143
- return { ...this.config };
144
- }
145
- /**
146
- * Check if a provider supports speculative decoding.
147
- */
148
- supportsSpeculativeDecoding(provider) {
149
- // Local providers (Ollama, vLLM) typically support it
150
- const localProviders = ['ollama', 'lmstudio', 'vllm'];
151
- if (localProviders.includes(provider))
152
- return true;
153
- // Check if provider has known speculative support
154
- const supportedProviders = ['together', 'fireworks', 'anyscale'];
155
- return supportedProviders.includes(provider);
156
- }
157
- /**
158
- * Get recommended speculation window based on model size.
159
- */
160
- getRecommendedWindow(modelSize) {
161
- switch (modelSize) {
162
- case 'small': return 6;
163
- case 'medium': return 4;
164
- case 'large': return 2;
165
- default: return 4;
166
- }
167
- }
168
- // ---- Private helpers ----
169
- registerDraftModels() {
170
- // Medusa-style models (multiple draft heads)
171
- this.draftModels.set('medusa', {
172
- id: 'medusa-7b',
173
- type: 'medusa',
174
- provider: 'auto',
175
- size: '7B equivalent',
176
- supportedWindows: [2, 4, 6, 8, 10],
177
- });
178
- // Lookahead decoding
179
- this.draftModels.set('lookahead', {
180
- id: 'lookahead-7b',
181
- type: 'lookahead',
182
- provider: 'auto',
183
- size: '7B equivalent',
184
- supportedWindows: [2, 4, 6],
185
- });
186
- // EAGLE decoding
187
- this.draftModels.set('eagle', {
188
- id: 'eagle-7b',
189
- type: 'eagle',
190
- provider: 'auto',
191
- size: '7B equivalent',
192
- supportedWindows: [2, 4, 6, 8],
193
- });
194
- // Spark (speculative decoding for transformers)
195
- this.draftModels.set('spark', {
196
- id: 'spark-7b',
197
- type: 'spark',
198
- provider: 'auto',
199
- size: '7B equivalent',
200
- supportedWindows: [2, 4, 6, 8, 10, 12],
201
- });
202
- }
203
- }
204
- exports.SpeculativeDecoding = SpeculativeDecoding;
205
- // ============================================================
206
- // Speculative Decoding Wrapper for LLM calls
207
- // ============================================================
208
- class SpeculativeDecodingWrapper {
209
- specDec;
210
- mainModelCall;
211
- constructor(mainModelCall, config) {
212
- this.mainModelCall = mainModelCall;
213
- this.specDec = new SpeculativeDecoding(config);
214
- }
215
- /**
216
- * Generate with speculative decoding.
217
- */
218
- async generate(prompt, maxTokens, options) {
219
- if (!this.specDec.shouldUse(options?.useSpecDec)) {
220
- // Fallback to normal generation
221
- const text = await this.mainModelCall(prompt, options);
222
- return {
223
- text,
224
- result: {
225
- used: false,
226
- draftTokens: 0,
227
- acceptedTokens: 0,
228
- acceptanceRate: 0,
229
- estimatedTimeSaved: 0,
230
- },
231
- };
232
- }
233
- // Speculative generation
234
- const startTime = Date.now();
235
- const draftTokens = await this.specDec.generateDraft(prompt, maxTokens);
236
- const verification = await this.specDec.verifyDraft(prompt, draftTokens.tokens);
237
- let text;
238
- if (verification.allAccepted) {
239
- // All draft tokens accepted, continue with main model
240
- text = verification.actualTokens.join('') +
241
- await this.mainModelCall(prompt, options);
242
- }
243
- else {
244
- // Some tokens rejected, need to rewind
245
- text = verification.actualTokens.join('');
246
- // Re-prompt with corrected context
247
- text += await this.mainModelCall(prompt + text, options);
248
- }
249
- const endTime = Date.now();
250
- const speedup = this.specDec.calculateSpeedup({
251
- used: true,
252
- draftTokens: draftTokens.tokens.length,
253
- acceptedTokens: verification.acceptedIndices.length,
254
- acceptanceRate: verification.acceptedIndices.length / draftTokens.tokens.length,
255
- estimatedTimeSaved: endTime - startTime,
256
- });
257
- return {
258
- text,
259
- result: {
260
- used: true,
261
- draftTokens: draftTokens.tokens.length,
262
- acceptedTokens: verification.acceptedIndices.length,
263
- acceptanceRate: verification.acceptedIndices.length / draftTokens.tokens.length,
264
- estimatedTimeSaved: Math.round((1 - 1 / speedup) * (endTime - startTime)),
265
- },
266
- };
267
- }
268
- }
269
- exports.SpeculativeDecodingWrapper = SpeculativeDecodingWrapper;
270
- // ============================================================
271
- // Factory
272
- // ============================================================
273
- function createSpeculativeDecoding(config) {
274
- return new SpeculativeDecoding(config);
275
- }
276
- //# sourceMappingURL=speculativeDecoding.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"speculativeDecoding.js","sourceRoot":"","sources":["../../src/inference/speculativeDecoding.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;;;;;GAgBG;;;AAgWH,8DAIC;AApSD,+DAA+D;AAC/D,iCAAiC;AACjC,+DAA+D;AAE/D,MAAa,mBAAmB;IACtB,MAAM,CAAoB;IAC1B,WAAW,GAA4B,IAAI,GAAG,EAAE,CAAC;IAEzD,YAAY,MAAmC;QAC7C,IAAI,CAAC,MAAM,GAAG;YACZ,OAAO,EAAE,MAAM,EAAE,OAAO,IAAI,KAAK;YACjC,cAAc,EAAE,MAAM,EAAE,cAAc,IAAI,QAAQ;YAClD,iBAAiB,EAAE,MAAM,EAAE,iBAAiB,IAAI,CAAC;YACjD,WAAW,EAAE,MAAM,EAAE,WAAW,IAAI,GAAG;YACvC,QAAQ,EAAE,MAAM,EAAE,QAAQ;YAC1B,KAAK,EAAE,MAAM,EAAE,KAAK;SACrB,CAAC;QAEF,kCAAkC;QAClC,IAAI,CAAC,mBAAmB,EAAE,CAAC;IAC7B,CAAC;IAED;;;OAGG;IACH,SAAS,CAAC,WAAqB;QAC7B,IAAI,WAAW;YAAE,OAAO,IAAI,CAAC;QAC7B,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO;YAAE,OAAO,KAAK,CAAC;QAEvC,qDAAqD;QACrD,gCAAgC;QAChC,OAAO,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC;IAC7B,CAAC;IAED;;OAEG;IACH,aAAa,CAAC,IAAqB;QACjC,MAAM,SAAS,GAAG,IAAI,IAAI,IAAI,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC;QAEjE,MAAM,KAAK,GAAG,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QAC9C,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,iCAAiC;YACjC,OAAO;gBACL,EAAE,EAAE,SAAS,SAAS,EAAE;gBACxB,IAAI,EAAE,SAAS;gBACf,QAAQ,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,IAAI,MAAM;gBACxC,IAAI,EAAE,OAAO;gBACb,gBAAgB,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;aAC/B,CAAC;QACJ,CAAC;QAED,OAAO,KAAK,CAAC;IACf,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,aAAa,CACjB,MAAc,EACd,SAAiB;QAEjB,MAAM,UAAU,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;QACxC,IAAI,CAAC,UAAU,EAAE,CAAC;YAChB,MAAM,IAAI,KAAK,CAAC,0BAA0B,CAAC,CAAC;QAC9C,CAAC;QAED,2DAA2D;QAC3D,8BAA8B;QAC9B,OAAO;YACL,MAAM,EAAE,EAAE;YACV,MAAM,EAAE,EAAE;SACX,CAAC;IACJ,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,WAAW,CACf,MAAc,EACd,WAAqB;QAErB,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC7B,OAAO;gBACL,WAAW,EAAE,IAAI;gBACjB,eAAe,EAAE,EAAE;gBACnB,eAAe,EAAE,EAAE;gBACnB,YAAY,EAAE,EAAE;gBAChB,WAAW,EAAE,CAAC;aACf,CAAC;QACJ,CAAC;QAED,0DAA0D;QAC1D,0CAA0C;QAC1C,OAAO;YACL,WAAW,EAAE,IAAI;YACjB,eAAe,EAAE,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;YAC7C,eAAe,EAAE,EAAE;YACnB,YAAY,EAAE,WAAW;YACzB,WAAW,EAAE,CAAC;SACf,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,MAAyB;QACxC,IAAI,CAAC,MAAM,CAAC,IAAI,IAAI,MAAM,CAAC,WAAW,KAAK,CAAC;YAAE,OAAO,GAAG,CAAC;QAEzD,+CAA+C;QAC/C,mCAAmC;QACnC,MAAM,cAAc,GAAG,MAAM,CAAC,cAAc,CAAC;QAE7C,8CAA8C;QAC9C,IAAI,cAAc,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;QACtC,IAAI,cAAc,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;QACtC,IAAI,cAAc,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;QACtC,IAAI,cAAc,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;QACtC,IAAI,cAAc,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;QACtC,OAAO,GAAG,CAAC;IACb,CAAC;IAED;;OAEG;IACH,YAAY,CAAC,OAAmC;QAC9C,IAAI,CAAC,MAAM,GAAG,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,OAAO,EAAE,CAAC;IAC/C,CAAC;IAED;;OAEG;IACH,SAAS;QACP,OAAO,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;IAC5B,CAAC;IAED;;OAEG;IACH,2BAA2B,CAAC,QAAgB;QAC1C,sDAAsD;QACtD,MAAM,cAAc,GAAG,CAAC,QAAQ,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC;QACtD,IAAI,cAAc,CAAC,QAAQ,CAAC,QAAQ,CAAC;YAAE,OAAO,IAAI,CAAC;QAEnD,kDAAkD;QAClD,MAAM,kBAAkB,GAAG,CAAC,UAAU,EAAE,WAAW,EAAE,UAAU,CAAC,CAAC;QACjE,OAAO,kBAAkB,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAC/C,CAAC;IAED;;OAEG;IACH,oBAAoB,CAAC,SAAuC;QAC1D,QAAQ,SAAS,EAAE,CAAC;YAClB,KAAK,OAAO,CAAC,CAAC,OAAO,CAAC,CAAC;YACvB,KAAK,QAAQ,CAAC,CAAC,OAAO,CAAC,CAAC;YACxB,KAAK,OAAO,CAAC,CAAC,OAAO,CAAC,CAAC;YACvB,OAAO,CAAC,CAAC,OAAO,CAAC,CAAC;QACpB,CAAC;IACH,CAAC;IAED,4BAA4B;IAEpB,mBAAmB;QACzB,6CAA6C;QAC7C,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,EAAE;YAC7B,EAAE,EAAE,WAAW;YACf,IAAI,EAAE,QAAQ;YACd,QAAQ,EAAE,MAAM;YAChB,IAAI,EAAE,eAAe;YACrB,gBAAgB,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;SACnC,CAAC,CAAC;QAEH,qBAAqB;QACrB,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,WAAW,EAAE;YAChC,EAAE,EAAE,cAAc;YAClB,IAAI,EAAE,WAAW;YACjB,QAAQ,EAAE,MAAM;YAChB,IAAI,EAAE,eAAe;YACrB,gBAAgB,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;SAC5B,CAAC,CAAC;QAEH,iBAAiB;QACjB,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,OAAO,EAAE;YAC5B,EAAE,EAAE,UAAU;YACd,IAAI,EAAE,OAAO;YACb,QAAQ,EAAE,MAAM;YAChB,IAAI,EAAE,eAAe;YACrB,gBAAgB,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;SAC/B,CAAC,CAAC;QAEH,gDAAgD;QAChD,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,OAAO,EAAE;YAC5B,EAAE,EAAE,UAAU;YACd,IAAI,EAAE,OAAO;YACb,QAAQ,EAAE,MAAM;YAChB,IAAI,EAAE,eAAe;YACrB,gBAAgB,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,EAAE,EAAE,CAAC;SACvC,CAAC,CAAC;IACL,CAAC;CACF;AAvMD,kDAuMC;AAED,+DAA+D;AAC/D,6CAA6C;AAC7C,+DAA+D;AAE/D,MAAa,0BAA0B;IAC7B,OAAO,CAAsB;IAC7B,aAAa,CAAqD;IAE1E,YACE,aAAiE,EACjE,MAAmC;QAEnC,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,OAAO,GAAG,IAAI,mBAAmB,CAAC,MAAM,CAAC,CAAC;IACjD,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,QAAQ,CACZ,MAAc,EACd,SAAiB,EACjB,OAAwD;QAExD,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,OAAO,EAAE,UAAU,CAAC,EAAE,CAAC;YACjD,gCAAgC;YAChC,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;YACvD,OAAO;gBACL,IAAI;gBACJ,MAAM,EAAE;oBACN,IAAI,EAAE,KAAK;oBACX,WAAW,EAAE,CAAC;oBACd,cAAc,EAAE,CAAC;oBACjB,cAAc,EAAE,CAAC;oBACjB,kBAAkB,EAAE,CAAC;iBACtB;aACF,CAAC;QACJ,CAAC;QAED,yBAAyB;QACzB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QACxE,MAAM,YAAY,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,MAAM,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;QAEhF,IAAI,IAAY,CAAC;QAEjB,IAAI,YAAY,CAAC,WAAW,EAAE,CAAC;YAC7B,sDAAsD;YACtD,IAAI,GAAG,YAAY,CAAC,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC;gBACvC,MAAM,IAAI,CAAC,aAAa,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAC9C,CAAC;aAAM,CAAC;YACN,uCAAuC;YACvC,IAAI,GAAG,YAAY,CAAC,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAC1C,mCAAmC;YACnC,IAAI,IAAI,MAAM,IAAI,CAAC,aAAa,CAAC,MAAM,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAC3D,CAAC;QAED,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC3B,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,gBAAgB,CAAC;YAC5C,IAAI,EAAE,IAAI;YACV,WAAW,EAAE,WAAW,CAAC,MAAM,CAAC,MAAM;YACtC,cAAc,EAAE,YAAY,CAAC,eAAe,CAAC,MAAM;YACnD,cAAc,EAAE,YAAY,CAAC,eAAe,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC,MAAM;YAC/E,kBAAkB,EAAE,OAAO,GAAG,SAAS;SACxC,CAAC,CAAC;QAEH,OAAO;YACL,IAAI;YACJ,MAAM,EAAE;gBACN,IAAI,EAAE,IAAI;gBACV,WAAW,EAAE,WAAW,CAAC,MAAM,CAAC,MAAM;gBACtC,cAAc,EAAE,YAAY,CAAC,eAAe,CAAC,MAAM;gBACnD,cAAc,EAAE,YAAY,CAAC,eAAe,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC,MAAM;gBAC/E,kBAAkB,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC,GAAG,CAAC,OAAO,GAAG,SAAS,CAAC,CAAC;aAC1E;SACF,CAAC;IACJ,CAAC;CACF;AAzED,gEAyEC;AAED,+DAA+D;AAC/D,UAAU;AACV,+DAA+D;AAE/D,SAAgB,yBAAyB,CACvC,MAAmC;IAEnC,OAAO,IAAI,mBAAmB,CAAC,MAAM,CAAC,CAAC;AACzC,CAAC"}
@@ -1,117 +0,0 @@
1
- /**
2
- * A3M Router - Provider Health Score
3
- *
4
- * Probabilistic health scoring (not binary up/down).
5
- * Uses latency history, error rate, partial failure patterns, queue depth estimates.
6
- *
7
- * Usage:
8
- * const healthMonitor = new ProviderHealthMonitor();
9
- * healthMonitor.recordLatency('groq', 250);
10
- * healthMonitor.recordError('groq', 'rate_limit');
11
- * const score = healthMonitor.getHealthScore('groq');
12
- * console.log(score); // { score: 0.85, confidence: 0.7, status: 'healthy' }
13
- */
14
- import { ProviderTier } from '../providers/providerConfig';
15
- export type HealthStatus = 'healthy' | 'degraded' | 'critical';
16
- export interface HealthScore {
17
- /** Overall health score 0-1 */
18
- score: number;
19
- /** Confidence in the score 0-1 */
20
- confidence: number;
21
- /** Human-readable status */
22
- status: HealthStatus;
23
- /** Component scores */
24
- components: {
25
- latency: number;
26
- errorRate: number;
27
- recentTrend: number;
28
- };
29
- /** Last updated timestamp */
30
- lastUpdated: number;
31
- /** Metadata about current state */
32
- metadata: {
33
- sampleSize: number;
34
- isAnomaly: boolean;
35
- anomalyReason?: string;
36
- };
37
- }
38
- export interface HealthRecord {
39
- timestamp: number;
40
- latency?: number;
41
- success?: boolean;
42
- errorType?: string;
43
- partialFailure?: boolean;
44
- }
45
- export interface HealthConfig {
46
- /** EWMA alpha for latency smoothing */
47
- latencyAlpha?: number;
48
- /** EWMA alpha for error rate smoothing */
49
- errorAlpha?: number;
50
- /** Window size for anomaly detection */
51
- anomalyWindow?: number;
52
- /** Standard deviation threshold for anomaly */
53
- anomalyThreshold?: number;
54
- /** Health score thresholds */
55
- healthyThreshold?: number;
56
- /** Degraded threshold */
57
- degradedThreshold?: number;
58
- }
59
- export declare class ProviderHealthMonitor {
60
- private records;
61
- private config;
62
- private ewmaLatency;
63
- private ewmaErrorRate;
64
- private anomalyScores;
65
- constructor(config?: HealthConfig);
66
- /**
67
- * Record a successful request with latency.
68
- */
69
- recordSuccess(providerId: string, latencyMs: number): void;
70
- /**
71
- * Record a failed request.
72
- */
73
- recordError(providerId: string, errorType?: string): void;
74
- /**
75
- * Record a partial failure (slow response, incomplete output, etc).
76
- */
77
- recordPartialFailure(providerId: string, latencyMs?: number): void;
78
- /**
79
- * Record queue depth estimate (when available from provider metadata).
80
- */
81
- recordQueueDepth(providerId: string, depth: number): void;
82
- /**
83
- * Get the current health score for a provider.
84
- */
85
- getHealthScore(providerId: string): HealthScore;
86
- /**
87
- * Get health scores for all tracked providers.
88
- */
89
- getAllHealthScores(): Record<string, HealthScore>;
90
- /**
91
- * Get the recommended provider based on health scores.
92
- * Excludes providers that are 'critical'.
93
- */
94
- getRecommendedProvider(providers: Array<{
95
- id: string;
96
- tier?: ProviderTier;
97
- priority?: number;
98
- }>): string | null;
99
- /**
100
- * Clear all records for a provider.
101
- */
102
- clearProvider(providerId: string): void;
103
- /**
104
- * Clear all records for all providers.
105
- */
106
- clearAll(): void;
107
- private addRecord;
108
- private getRecords;
109
- private updateEwmaLatency;
110
- private updateEwmaErrorRate;
111
- private updateAnomalyScore;
112
- private calculateLatencyScore;
113
- private calculateErrorScore;
114
- private calculateTrendScore;
115
- }
116
- export declare function createProviderHealthMonitor(config?: HealthConfig): ProviderHealthMonitor;
117
- //# sourceMappingURL=providerHealth.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"providerHealth.d.ts","sourceRoot":"","sources":["../../src/providers/providerHealth.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,6BAA6B,CAAC;AAM3D,MAAM,MAAM,YAAY,GAAG,SAAS,GAAG,UAAU,GAAG,UAAU,CAAC;AAE/D,MAAM,WAAW,WAAW;IAC1B,+BAA+B;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,kCAAkC;IAClC,UAAU,EAAE,MAAM,CAAC;IACnB,4BAA4B;IAC5B,MAAM,EAAE,YAAY,CAAC;IACrB,uBAAuB;IACvB,UAAU,EAAE;QACV,OAAO,EAAE,MAAM,CAAC;QAChB,SAAS,EAAE,MAAM,CAAC;QAClB,WAAW,EAAE,MAAM,CAAC;KACrB,CAAC;IACF,6BAA6B;IAC7B,WAAW,EAAE,MAAM,CAAC;IACpB,mCAAmC;IACnC,QAAQ,EAAE;QACR,UAAU,EAAE,MAAM,CAAC;QACnB,SAAS,EAAE,OAAO,CAAC;QACnB,aAAa,CAAC,EAAE,MAAM,CAAC;KACxB,CAAC;CACH;AAED,MAAM,WAAW,YAAY;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,cAAc,CAAC,EAAE,OAAO,CAAC;CAC1B;AAED,MAAM,WAAW,YAAY;IAC3B,uCAAuC;IACvC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,0CAA0C;IAC1C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,wCAAwC;IACxC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,+CAA+C;IAC/C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,8BAA8B;IAC9B,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,yBAAyB;IACzB,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAeD,qBAAa,qBAAqB;IAChC,OAAO,CAAC,OAAO,CAA0C;IACzD,OAAO,CAAC,MAAM,CAAyB;IACvC,OAAO,CAAC,WAAW,CAAkC;IACrD,OAAO,CAAC,aAAa,CAAkC;IACvD,OAAO,CAAC,aAAa,CAAkC;gBAE3C,MAAM,GAAE,YAAiB;IAIrC;;OAEG;IACH,aAAa,CAAC,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI;IAW1D;;OAEG;IACH,WAAW,CAAC,UAAU,EAAE,MAAM,EAAE,SAAS,GAAE,MAAkB,GAAG,IAAI;IASpE;;OAEG;IACH,oBAAoB,CAAC,UAAU,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI;IAelE;;OAEG;IACH,gBAAgB,CAAC,UAAU,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,IAAI;IAWzD;;OAEG;IACH,cAAc,CAAC,UAAU,EAAE,MAAM,GAAG,WAAW;IAoE/C;;OAEG;IACH,kBAAkB,IAAI,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC;IAQjD;;;OAGG;IACH,sBAAsB,CACpB,SAAS,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,YAAY,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,GACvE,MAAM,GAAG,IAAI;IAqBhB;;OAEG;IACH,aAAa,CAAC,UAAU,EAAE,MAAM,GAAG,IAAI;IAOvC;;OAEG;IACH,QAAQ,IAAI,IAAI;IAShB,OAAO,CAAC,SAAS;IAajB,OAAO,CAAC,UAAU;IAIlB,OAAO,CAAC,iBAAiB;IAYzB,OAAO,CAAC,mBAAmB;IAa3B,OAAO,CAAC,kBAAkB;IA2B1B,OAAO,CAAC,qBAAqB;IAc7B,OAAO,CAAC,mBAAmB;IAM3B,OAAO,CAAC,mBAAmB;CAkB5B;AAMD,wBAAgB,2BAA2B,CAAC,MAAM,CAAC,EAAE,YAAY,GAAG,qBAAqB,CAExF"}