@llm4ts/core 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 +47 -0
- package/dist/AgentSession.d.ts +66 -0
- package/dist/AgentSession.d.ts.map +1 -0
- package/dist/AgentSession.js +53 -0
- package/dist/AgentSession.js.map +1 -0
- package/dist/Capability.d.ts +101 -0
- package/dist/Capability.d.ts.map +1 -0
- package/dist/Capability.js +210 -0
- package/dist/Capability.js.map +1 -0
- package/dist/Connector.d.ts +69 -0
- package/dist/Connector.d.ts.map +1 -0
- package/dist/Connector.js +86 -0
- package/dist/Connector.js.map +1 -0
- package/dist/ConnectorConfig.d.ts +70 -0
- package/dist/ConnectorConfig.d.ts.map +1 -0
- package/dist/ConnectorConfig.js +110 -0
- package/dist/ConnectorConfig.js.map +1 -0
- package/dist/ConnectorRegistry.d.ts +28 -0
- package/dist/ConnectorRegistry.d.ts.map +1 -0
- package/dist/ConnectorRegistry.js +87 -0
- package/dist/ConnectorRegistry.js.map +1 -0
- package/dist/ContextManagement.d.ts +78 -0
- package/dist/ContextManagement.d.ts.map +1 -0
- package/dist/ContextManagement.js +221 -0
- package/dist/ContextManagement.js.map +1 -0
- package/dist/Conversation.d.ts +18 -0
- package/dist/Conversation.d.ts.map +1 -0
- package/dist/Conversation.js +17 -0
- package/dist/Conversation.js.map +1 -0
- package/dist/Errors.d.ts +64 -0
- package/dist/Errors.d.ts.map +1 -0
- package/dist/Errors.js +76 -0
- package/dist/Errors.js.map +1 -0
- package/dist/HttpClient.d.ts +43 -0
- package/dist/HttpClient.d.ts.map +1 -0
- package/dist/HttpClient.js +99 -0
- package/dist/HttpClient.js.map +1 -0
- package/dist/LlmService.d.ts +24 -0
- package/dist/LlmService.d.ts.map +1 -0
- package/dist/LlmService.js +4 -0
- package/dist/LlmService.js.map +1 -0
- package/dist/Models.d.ts +138 -0
- package/dist/Models.d.ts.map +1 -0
- package/dist/Models.js +209 -0
- package/dist/Models.js.map +1 -0
- package/dist/ProcessExecutor.d.ts +57 -0
- package/dist/ProcessExecutor.d.ts.map +1 -0
- package/dist/ProcessExecutor.js +71 -0
- package/dist/ProcessExecutor.js.map +1 -0
- package/dist/RateLimiter.d.ts +44 -0
- package/dist/RateLimiter.d.ts.map +1 -0
- package/dist/RateLimiter.js +155 -0
- package/dist/RateLimiter.js.map +1 -0
- package/dist/Streaming.d.ts +17 -0
- package/dist/Streaming.d.ts.map +1 -0
- package/dist/Streaming.js +65 -0
- package/dist/Streaming.js.map +1 -0
- package/dist/StructuredOutput.d.ts +10 -0
- package/dist/StructuredOutput.d.ts.map +1 -0
- package/dist/StructuredOutput.js +97 -0
- package/dist/StructuredOutput.js.map +1 -0
- package/dist/TemporaryFiles.d.ts +20 -0
- package/dist/TemporaryFiles.d.ts.map +1 -0
- package/dist/TemporaryFiles.js +18 -0
- package/dist/TemporaryFiles.js.map +1 -0
- package/dist/UsageLimits.d.ts +5 -0
- package/dist/UsageLimits.d.ts.map +1 -0
- package/dist/UsageLimits.js +187 -0
- package/dist/UsageLimits.js.map +1 -0
- package/dist/eval/Checks.d.ts +6 -0
- package/dist/eval/Checks.d.ts.map +1 -0
- package/dist/eval/Checks.js +39 -0
- package/dist/eval/Checks.js.map +1 -0
- package/dist/eval/Eval.d.ts +57 -0
- package/dist/eval/Eval.d.ts.map +1 -0
- package/dist/eval/Eval.js +91 -0
- package/dist/eval/Eval.js.map +1 -0
- package/dist/eval/EvalSuite.d.ts +21 -0
- package/dist/eval/EvalSuite.d.ts.map +1 -0
- package/dist/eval/EvalSuite.js +30 -0
- package/dist/eval/EvalSuite.js.map +1 -0
- package/dist/eval/Evaluator.d.ts +10 -0
- package/dist/eval/Evaluator.d.ts.map +1 -0
- package/dist/eval/Evaluator.js +14 -0
- package/dist/eval/Evaluator.js.map +1 -0
- package/dist/eval/Judge.d.ts +15 -0
- package/dist/eval/Judge.d.ts.map +1 -0
- package/dist/eval/Judge.js +73 -0
- package/dist/eval/Judge.js.map +1 -0
- package/dist/observability/MeteredLlmService.d.ts +16 -0
- package/dist/observability/MeteredLlmService.d.ts.map +1 -0
- package/dist/observability/MeteredLlmService.js +66 -0
- package/dist/observability/MeteredLlmService.js.map +1 -0
- package/dist/observability/Metrics.d.ts +71 -0
- package/dist/observability/Metrics.d.ts.map +1 -0
- package/dist/observability/Metrics.js +147 -0
- package/dist/observability/Metrics.js.map +1 -0
- package/dist/observability/Redaction.d.ts +8 -0
- package/dist/observability/Redaction.d.ts.map +1 -0
- package/dist/observability/Redaction.js +38 -0
- package/dist/observability/Redaction.js.map +1 -0
- package/dist/observability/StreamRecorder.d.ts +9 -0
- package/dist/observability/StreamRecorder.d.ts.map +1 -0
- package/dist/observability/StreamRecorder.js +10 -0
- package/dist/observability/StreamRecorder.js.map +1 -0
- package/dist/observability/StructuredLogger.d.ts +31 -0
- package/dist/observability/StructuredLogger.d.ts.map +1 -0
- package/dist/observability/StructuredLogger.js +34 -0
- package/dist/observability/StructuredLogger.js.map +1 -0
- package/dist/observability/Tracing.d.ts +38 -0
- package/dist/observability/Tracing.d.ts.map +1 -0
- package/dist/observability/Tracing.js +75 -0
- package/dist/observability/Tracing.js.map +1 -0
- package/dist/providers/AnthropicModels.d.ts +95 -0
- package/dist/providers/AnthropicModels.d.ts.map +1 -0
- package/dist/providers/AnthropicModels.js +83 -0
- package/dist/providers/AnthropicModels.js.map +1 -0
- package/dist/providers/AnthropicProvider.d.ts +13 -0
- package/dist/providers/AnthropicProvider.d.ts.map +1 -0
- package/dist/providers/AnthropicProvider.js +177 -0
- package/dist/providers/AnthropicProvider.js.map +1 -0
- package/dist/providers/AntigravityConnector.d.ts +7 -0
- package/dist/providers/AntigravityConnector.d.ts.map +1 -0
- package/dist/providers/AntigravityConnector.js +52 -0
- package/dist/providers/AntigravityConnector.js.map +1 -0
- package/dist/providers/ClaudeAgentSession.d.ts +11 -0
- package/dist/providers/ClaudeAgentSession.d.ts.map +1 -0
- package/dist/providers/ClaudeAgentSession.js +158 -0
- package/dist/providers/ClaudeAgentSession.js.map +1 -0
- package/dist/providers/ClaudeCliConnector.d.ts +9 -0
- package/dist/providers/ClaudeCliConnector.d.ts.map +1 -0
- package/dist/providers/ClaudeCliConnector.js +123 -0
- package/dist/providers/ClaudeCliConnector.js.map +1 -0
- package/dist/providers/CliSupport.d.ts +21 -0
- package/dist/providers/CliSupport.d.ts.map +1 -0
- package/dist/providers/CliSupport.js +62 -0
- package/dist/providers/CliSupport.js.map +1 -0
- package/dist/providers/CodexConnector.d.ts +12 -0
- package/dist/providers/CodexConnector.d.ts.map +1 -0
- package/dist/providers/CodexConnector.js +175 -0
- package/dist/providers/CodexConnector.js.map +1 -0
- package/dist/providers/ConnectorFactories.d.ts +15 -0
- package/dist/providers/ConnectorFactories.d.ts.map +1 -0
- package/dist/providers/ConnectorFactories.js +74 -0
- package/dist/providers/ConnectorFactories.js.map +1 -0
- package/dist/providers/CopilotConnector.d.ts +6 -0
- package/dist/providers/CopilotConnector.d.ts.map +1 -0
- package/dist/providers/CopilotConnector.js +35 -0
- package/dist/providers/CopilotConnector.js.map +1 -0
- package/dist/providers/CursorConnector.d.ts +8 -0
- package/dist/providers/CursorConnector.d.ts.map +1 -0
- package/dist/providers/CursorConnector.js +82 -0
- package/dist/providers/CursorConnector.js.map +1 -0
- package/dist/providers/GeminiApiProvider.d.ts +9 -0
- package/dist/providers/GeminiApiProvider.d.ts.map +1 -0
- package/dist/providers/GeminiApiProvider.js +181 -0
- package/dist/providers/GeminiApiProvider.js.map +1 -0
- package/dist/providers/GeminiCliProvider.d.ts +106 -0
- package/dist/providers/GeminiCliProvider.d.ts.map +1 -0
- package/dist/providers/GeminiCliProvider.js +596 -0
- package/dist/providers/GeminiCliProvider.js.map +1 -0
- package/dist/providers/GeminiModels.d.ts +92 -0
- package/dist/providers/GeminiModels.d.ts.map +1 -0
- package/dist/providers/GeminiModels.js +80 -0
- package/dist/providers/GeminiModels.js.map +1 -0
- package/dist/providers/GrokCliConnector.d.ts +8 -0
- package/dist/providers/GrokCliConnector.d.ts.map +1 -0
- package/dist/providers/GrokCliConnector.js +82 -0
- package/dist/providers/GrokCliConnector.js.map +1 -0
- package/dist/providers/LmStudioModels.d.ts +114 -0
- package/dist/providers/LmStudioModels.d.ts.map +1 -0
- package/dist/providers/LmStudioModels.js +102 -0
- package/dist/providers/LmStudioModels.js.map +1 -0
- package/dist/providers/LmStudioProvider.d.ts +13 -0
- package/dist/providers/LmStudioProvider.d.ts.map +1 -0
- package/dist/providers/LmStudioProvider.js +158 -0
- package/dist/providers/LmStudioProvider.js.map +1 -0
- package/dist/providers/MockProvider.d.ts +6 -0
- package/dist/providers/MockProvider.d.ts.map +1 -0
- package/dist/providers/MockProvider.js +271 -0
- package/dist/providers/MockProvider.js.map +1 -0
- package/dist/providers/OllamaModels.d.ts +68 -0
- package/dist/providers/OllamaModels.d.ts.map +1 -0
- package/dist/providers/OllamaModels.js +59 -0
- package/dist/providers/OllamaModels.js.map +1 -0
- package/dist/providers/OllamaProvider.d.ts +7 -0
- package/dist/providers/OllamaProvider.d.ts.map +1 -0
- package/dist/providers/OllamaProvider.js +134 -0
- package/dist/providers/OllamaProvider.js.map +1 -0
- package/dist/providers/OpenAIModels.d.ts +139 -0
- package/dist/providers/OpenAIModels.d.ts.map +1 -0
- package/dist/providers/OpenAIModels.js +123 -0
- package/dist/providers/OpenAIModels.js.map +1 -0
- package/dist/providers/OpenAIProvider.d.ts +9 -0
- package/dist/providers/OpenAIProvider.d.ts.map +1 -0
- package/dist/providers/OpenAIProvider.js +185 -0
- package/dist/providers/OpenAIProvider.js.map +1 -0
- package/dist/providers/OpenCodeCliConnector.d.ts +8 -0
- package/dist/providers/OpenCodeCliConnector.d.ts.map +1 -0
- package/dist/providers/OpenCodeCliConnector.js +93 -0
- package/dist/providers/OpenCodeCliConnector.js.map +1 -0
- package/dist/providers/OpenCodeModels.d.ts +61 -0
- package/dist/providers/OpenCodeModels.d.ts.map +1 -0
- package/dist/providers/OpenCodeModels.js +53 -0
- package/dist/providers/OpenCodeModels.js.map +1 -0
- package/dist/providers/OpenCodeProvider.d.ts +10 -0
- package/dist/providers/OpenCodeProvider.d.ts.map +1 -0
- package/dist/providers/OpenCodeProvider.js +154 -0
- package/dist/providers/OpenCodeProvider.js.map +1 -0
- package/dist/providers/PiConnector.d.ts +8 -0
- package/dist/providers/PiConnector.d.ts.map +1 -0
- package/dist/providers/PiConnector.js +99 -0
- package/dist/providers/PiConnector.js.map +1 -0
- package/dist/tools/Tool.d.ts +83 -0
- package/dist/tools/Tool.d.ts.map +1 -0
- package/dist/tools/Tool.js +278 -0
- package/dist/tools/Tool.js.map +1 -0
- package/dist/tools/ToolCallingExecutor.d.ts +18 -0
- package/dist/tools/ToolCallingExecutor.d.ts.map +1 -0
- package/dist/tools/ToolCallingExecutor.js +64 -0
- package/dist/tools/ToolCallingExecutor.js.map +1 -0
- package/dist/tools/ToolRegistry.d.ts +45 -0
- package/dist/tools/ToolRegistry.d.ts.map +1 -0
- package/dist/tools/ToolRegistry.js +133 -0
- package/dist/tools/ToolRegistry.js.map +1 -0
- package/package.json +105 -0
- package/src/.gitkeep +1 -0
- package/src/AgentSession.ts +69 -0
- package/src/Capability.ts +240 -0
- package/src/Connector.ts +220 -0
- package/src/ConnectorConfig.ts +146 -0
- package/src/ConnectorRegistry.ts +184 -0
- package/src/ContextManagement.ts +372 -0
- package/src/Conversation.ts +21 -0
- package/src/Errors.ts +87 -0
- package/src/HttpClient.ts +202 -0
- package/src/LlmService.ts +45 -0
- package/src/Models.ts +250 -0
- package/src/ProcessExecutor.ts +157 -0
- package/src/RateLimiter.ts +238 -0
- package/src/Streaming.ts +164 -0
- package/src/StructuredOutput.ts +138 -0
- package/src/TemporaryFiles.ts +51 -0
- package/src/UsageLimits.ts +242 -0
- package/src/eval/Checks.ts +68 -0
- package/src/eval/Eval.ts +112 -0
- package/src/eval/EvalSuite.ts +63 -0
- package/src/eval/Evaluator.ts +31 -0
- package/src/eval/Judge.ts +92 -0
- package/src/observability/MeteredLlmService.ts +156 -0
- package/src/observability/Metrics.ts +195 -0
- package/src/observability/Redaction.ts +55 -0
- package/src/observability/StreamRecorder.ts +32 -0
- package/src/observability/StructuredLogger.ts +62 -0
- package/src/observability/Tracing.ts +141 -0
- package/src/providers/AnthropicModels.ts +96 -0
- package/src/providers/AnthropicProvider.ts +320 -0
- package/src/providers/AntigravityConnector.ts +77 -0
- package/src/providers/ClaudeAgentSession.ts +226 -0
- package/src/providers/ClaudeCliConnector.ts +201 -0
- package/src/providers/CliSupport.ts +100 -0
- package/src/providers/CodexConnector.ts +251 -0
- package/src/providers/ConnectorFactories.ts +142 -0
- package/src/providers/CopilotConnector.ts +47 -0
- package/src/providers/CursorConnector.ts +120 -0
- package/src/providers/GeminiApiProvider.ts +337 -0
- package/src/providers/GeminiCliProvider.ts +793 -0
- package/src/providers/GeminiModels.ts +97 -0
- package/src/providers/GrokCliConnector.ts +117 -0
- package/src/providers/LmStudioModels.ts +117 -0
- package/src/providers/LmStudioProvider.ts +260 -0
- package/src/providers/MockProvider.ts +322 -0
- package/src/providers/OllamaModels.ts +64 -0
- package/src/providers/OllamaProvider.ts +236 -0
- package/src/providers/OpenAIModels.ts +152 -0
- package/src/providers/OpenAIProvider.ts +331 -0
- package/src/providers/OpenCodeCliConnector.ts +135 -0
- package/src/providers/OpenCodeModels.ts +60 -0
- package/src/providers/OpenCodeProvider.ts +278 -0
- package/src/providers/PiConnector.ts +142 -0
- package/src/tools/Tool.ts +379 -0
- package/src/tools/ToolCallingExecutor.ts +106 -0
- package/src/tools/ToolRegistry.ts +224 -0
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"UsageLimits.js","sourceRoot":"","sources":["../src/UsageLimits.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,QAAQ,MAAM,iBAAiB,CAAA;AAC3C,OAAO,KAAK,QAAQ,MAAM,iBAAiB,CAAA;AAC3C,OAAO,EAAE,cAAc,EAAE,eAAe,EAAiB,MAAM,aAAa,CAAA;AAE5E,MAAM,OAAO,GAAG,yCAAyC,CAAA;AACzD,MAAM,QAAQ,GAAG,wEAAwE,CAAA;AACzF,MAAM,WAAW,GAAG,wCAAwC,CAAA;AAC5D,MAAM,cAAc,GAAG,oBAAoB,CAAA;AAC3C,MAAM,eAAe,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;AAE7C,MAAM,kBAAkB,GAA0B;IAChD,yBAAyB;IACzB,OAAO;IACP,oBAAoB;IACpB,YAAY;IACZ,YAAY;IACZ,KAAK;CACN,CAAA;AAED,MAAM,mBAAmB,GAA0B;IACjD,GAAG,kBAAkB;IACrB,aAAa;IACb,gBAAgB;IAChB,mBAAmB;CACpB,CAAA;AAED,MAAM,cAAc,GAAG,CAAC,IAAY,EAAE,OAA8B,EAAW,EAAE;IAC/E,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,CAAA;IACrC,OAAO,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAA;AAC9D,CAAC,CAAA;AAED,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,IAAY,EAAW,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,kBAAkB,CAAC,CAAA;AAEhG,MAAM,cAAc,GAAG,CAAC,IAAY,EAAW,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,mBAAmB,CAAC,CAAA;AAE3F,MAAM,aAAa,GAAG,CAAC,aAAqB,EAAqB,EAAE;IACjE,IAAI,YAAY,GAAG,CAAC,CAAA;IAEpB,KAAK,MAAM,KAAK,IAAI,aAAa,CAAC,QAAQ,CAAC,cAAc,CAAC,EAAE,CAAC;QAC3D,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAA;QACtC,MAAM,IAAI,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,WAAW,EAAE,CAAA;QAE7C,QAAQ,IAAI,EAAE,CAAC;YACb,KAAK,GAAG;gBACN,YAAY,IAAI,MAAM,GAAG,UAAU,CAAA;gBACnC,MAAK;YACP,KAAK,GAAG;gBACN,YAAY,IAAI,MAAM,GAAG,SAAS,CAAA;gBAClC,MAAK;YACP,KAAK,GAAG;gBACN,YAAY,IAAI,MAAM,GAAG,MAAM,CAAA;gBAC/B,MAAK;YACP,KAAK,GAAG;gBACN,YAAY,IAAI,MAAM,GAAG,KAAK,CAAA;gBAC9B,MAAK;YACP;gBACE,YAAY,IAAI,MAAM,CAAA;QAC1B,CAAC;IACH,CAAC;IAED,OAAO,QAAQ,CAAC,MAAM,CAAC,YAAY,CAAC,CAAA;AACtC,CAAC,CAAA;AAOD,MAAM,kBAAkB,GAAG,CAAC,GAAW,EAA6B,EAAE;IACpE,MAAM,KAAK,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,qCAAqC,CAAC,CAAA;IACrE,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;QACnB,OAAO,SAAS,CAAA;IAClB,CAAC;IAED,MAAM,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAA;IAClC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAA;IACtC,MAAM,MAAM,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAA;IAC7C,IAAI,SAAS,GAAG,CAAC,IAAI,SAAS,GAAG,EAAE,IAAI,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,EAAE,EAAE,CAAC;QACjE,OAAO,SAAS,CAAA;IAClB,CAAC;IAED,OAAO;QACL,IAAI,EAAE,CAAC,SAAS,GAAG,EAAE,CAAC,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAClD,MAAM;KACP,CAAA;AACH,CAAC,CAAA;AAQD,MAAM,iBAAiB,GAAG,CAAC,QAAgB,EAAuB,EAAE,CAClE,IAAI,IAAI,CAAC,cAAc,CAAC,iBAAiB,EAAE;IACzC,QAAQ;IACR,IAAI,EAAE,SAAS;IACf,KAAK,EAAE,SAAS;IAChB,GAAG,EAAE,SAAS;IACd,IAAI,EAAE,SAAS;IACf,MAAM,EAAE,SAAS;IACjB,MAAM,EAAE,SAAS;IACjB,SAAS,EAAE,KAAK;CACjB,CAAC,CAAA;AAEJ,MAAM,WAAW,GAAG,CAClB,KAA6C,EAC7C,IAAkC,EAC1B,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,EAAE,KAAK,IAAI,GAAG,CAAC,CAAA;AAE3E,MAAM,UAAU,GAAG,CAAC,WAAmB,EAAE,QAAgB,EAAE,EAAE;IAC3D,MAAM,KAAK,GAAG,iBAAiB,CAAC,QAAQ,CAAC,CAAC,aAAa,CAAC,IAAI,IAAI,CAAC,WAAW,CAAC,CAAC,CAAA;IAC9E,OAAO;QACL,IAAI,EAAE,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC;QAChC,KAAK,EAAE,WAAW,CAAC,KAAK,EAAE,OAAO,CAAC;QAClC,GAAG,EAAE,WAAW,CAAC,KAAK,EAAE,KAAK,CAAC;QAC9B,IAAI,EAAE,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC;QAChC,MAAM,EAAE,WAAW,CAAC,KAAK,EAAE,QAAQ,CAAC;QACpC,MAAM,EAAE,WAAW,CAAC,KAAK,EAAE,QAAQ,CAAC;KACrC,CAAA;AACH,CAAC,CAAA;AAED,MAAM,oBAAoB,GAAG,CAAC,WAAmB,EAAE,QAAgB,EAAU,EAAE;IAC7E,MAAM,KAAK,GAAG,UAAU,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAA;IAC/C,MAAM,gBAAgB,GAAG,IAAI,CAAC,GAAG,CAC/B,KAAK,CAAC,IAAI,EACV,KAAK,CAAC,KAAK,GAAG,CAAC,EACf,KAAK,CAAC,GAAG,EACT,KAAK,CAAC,IAAI,EACV,KAAK,CAAC,MAAM,EACZ,KAAK,CAAC,MAAM,CACb,CAAA;IACD,OAAO,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,WAAW,GAAG,KAAK,CAAC,GAAG,KAAK,CAAA;AACnE,CAAC,CAAA;AAED,MAAM,iBAAiB,GAAG,CAAC,IAAkB,EAAE,IAAmB,EAAE,QAAgB,EAAU,EAAE;IAC9F,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,IAAI,CAAC,GAAG,EAAE,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC,CAAA;IACtF,MAAM,cAAc,GAAG,QAAQ,GAAG,oBAAoB,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAA;IAC1E,OAAO,QAAQ,GAAG,oBAAoB,CAAC,cAAc,EAAE,QAAQ,CAAC,CAAA;AAClE,CAAC,CAAA;AAED,MAAM,gBAAgB,GAAG,CAAC,IAAkB,EAAgB,EAAE;IAC5D,MAAM,IAAI,GAAG,IAAI,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,IAAI,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAA;IACxE,OAAO;QACL,IAAI,EAAE,IAAI,CAAC,cAAc,EAAE;QAC3B,KAAK,EAAE,IAAI,CAAC,WAAW,EAAE,GAAG,CAAC;QAC7B,GAAG,EAAE,IAAI,CAAC,UAAU,EAAE;KACvB,CAAA;AACH,CAAC,CAAA;AAED,MAAM,cAAc,GAAG,CACrB,IAAY,EACZ,OAAe,EACf,GAAiB,EACjB,QAAgB,EACU,EAAE;IAC5B,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC,CAAA;IACxC,MAAM,IAAI,GAAG,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAC,OAAO,CAAC,CAAA;IAC5E,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;QACvB,OAAO,SAAS,CAAA;IAClB,CAAC;IAED,IAAI,CAAC;QACH,MAAM,SAAS,GAAG,QAAQ,CAAC,aAAa,CAAC,GAAG,CAAC,CAAA;QAC7C,MAAM,QAAQ,GAAG,UAAU,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAA;QAChD,MAAM,KAAK,GAAiB;YAC1B,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,KAAK,EAAE,QAAQ,CAAC,KAAK;YACrB,GAAG,EAAE,QAAQ,CAAC,GAAG;SAClB,CAAA;QACD,MAAM,cAAc,GAAG,iBAAiB,CAAC,KAAK,EAAE,IAAI,EAAE,QAAQ,CAAC,CAAA;QAC/D,MAAM,WAAW,GACf,cAAc,GAAG,SAAS;YACxB,CAAC,CAAC,cAAc;YAChB,CAAC,CAAC,iBAAiB,CAAC,gBAAgB,CAAC,KAAK,CAAC,EAAE,IAAI,EAAE,QAAQ,CAAC,CAAA;QAEhE,OAAO,QAAQ,CAAC,UAAU,CAAC,WAAW,CAAC,CAAA;IACzC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAA;IAClB,CAAC;AACH,CAAC,CAAA;AAED,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAChC,QAAgB,EAChB,IAAY,EACZ,GAAiB,EACjB,QAAgB,EACM,EAAE;IACxB,QAAQ,QAAQ,EAAE,CAAC;QACjB,KAAK,OAAO,CAAC,CAAC,CAAC;YACb,MAAM,OAAO,GAAG,cAAc,CAAC,IAAI,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,CAAC,CAAA;YAC5D,OAAO,OAAO,KAAK,SAAS;gBAC1B,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,eAAe,CAAC,IAAI,CAAC;oBACnB,OAAO;oBACP,QAAQ;oBACR,OAAO,EAAE,IAAI;iBACd,CAAC,CAAA;QACR,CAAC;QACD,KAAK,QAAQ,CAAC,CAAC,CAAC;YACd,MAAM,OAAO,GAAG,cAAc,CAAC,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,QAAQ,CAAC,CAAA;YAC7D,OAAO,OAAO,KAAK,SAAS;gBAC1B,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,eAAe,CAAC,IAAI,CAAC;oBACnB,OAAO;oBACP,QAAQ;oBACR,OAAO,EAAE,IAAI;iBACd,CAAC,CAAA;QACR,CAAC;QACD,KAAK,QAAQ,CAAC,CAAC,CAAC;YACd,MAAM,kBAAkB,GAAG,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC,CAAC,CAAA;YACvD,IAAI,kBAAkB,KAAK,SAAS,EAAE,CAAC;gBACrC,MAAM,QAAQ,GAAG,aAAa,CAAC,kBAAkB,CAAC,CAAA;gBAClD,OAAO,QAAQ,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,QAAQ,CAAC,QAAQ,CAAC,eAAe,CAAC;oBACtE,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;oBAC/C,CAAC,CAAC,eAAe,CAAC,IAAI,CAAC;wBACnB,OAAO,EAAE,QAAQ,CAAC,WAAW,CAAC,GAAG,EAAE,QAAQ,CAAC;wBAC5C,QAAQ;wBACR,OAAO,EAAE,IAAI;qBACd,CAAC,CAAA;YACR,CAAC;YAED,OAAO,aAAa,CAAC,IAAI,CAAC;gBACxB,CAAC,CAAC,eAAe,CAAC,IAAI,CAAC;oBACnB,QAAQ;oBACR,OAAO,EAAE,IAAI;iBACd,CAAC;gBACJ,CAAC,CAAC,SAAS,CAAA;QACf,CAAC;QACD,KAAK,MAAM,CAAC;QACZ,KAAK,QAAQ,CAAC;QACd,KAAK,UAAU;YACb,OAAO,cAAc,CAAC,IAAI,CAAC;gBACzB,CAAC,CAAC,eAAe,CAAC,IAAI,CAAC;oBACnB,QAAQ;oBACR,OAAO,EAAE,IAAI;iBACd,CAAC;gBACJ,CAAC,CAAC,SAAS,CAAA;QACf;YACE,OAAO,SAAS,CAAA;IACpB,CAAC;AACH,CAAC,CAAA"}
|
|
@@ -0,0 +1,6 @@
|
|
|
1
|
+
import { type Evaluator } from "./Evaluator.ts";
|
|
2
|
+
export declare const noPii: (maxScore?: number) => Evaluator<string>;
|
|
3
|
+
export declare const validJson: (maxScore?: number) => Evaluator<string>;
|
|
4
|
+
export declare const matches: (regex: string, name?: string, maxScore?: number) => Evaluator<string>;
|
|
5
|
+
export declare const lengthBetween: (min: number, max: number, maxScore?: number) => Evaluator<string>;
|
|
6
|
+
//# sourceMappingURL=Checks.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Checks.d.ts","sourceRoot":"","sources":["../../src/eval/Checks.ts"],"names":[],"mappings":"AAEA,OAAO,EAAiB,KAAK,SAAS,EAAE,MAAM,gBAAgB,CAAA;AAqB9D,eAAO,MAAM,KAAK,GAAI,iBAAY,KAAG,SAAS,CAAC,MAAM,CAWjD,CAAA;AAEJ,eAAO,MAAM,SAAS,GAAI,iBAAY,KAAG,SAAS,CAAC,MAAM,CAWrD,CAAA;AAEJ,eAAO,MAAM,OAAO,GAAI,OAAO,MAAM,EAAE,aAAe,EAAE,iBAAY,KAAG,SAAS,CAAC,MAAM,CAMnF,CAAA;AAEJ,eAAO,MAAM,aAAa,GAAI,KAAK,MAAM,EAAE,KAAK,MAAM,EAAE,iBAAY,KAAG,SAAS,CAAC,MAAM,CAUpF,CAAA"}
|
|
@@ -0,0 +1,39 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import { DimensionScore, EvalResult } from "./Eval.js";
|
|
3
|
+
import { makeEvaluator } from "./Evaluator.js";
|
|
4
|
+
const piiPatterns = [
|
|
5
|
+
["email", /[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}/],
|
|
6
|
+
["ssn", /\b\d{3}-\d{2}-\d{4}\b/],
|
|
7
|
+
["credit-card", /\b(?:\d[ -]?){13,16}\b/],
|
|
8
|
+
["phone", /\b\d{3}[-.\s]?\d{3}[-.\s]?\d{4}\b/],
|
|
9
|
+
["ipv4", /\b(?:\d{1,3}\.){3}\d{1,3}\b/]
|
|
10
|
+
];
|
|
11
|
+
const one = (name, ok, maxScore, why) => EvalResult.make({
|
|
12
|
+
scores: [
|
|
13
|
+
DimensionScore.make({
|
|
14
|
+
name,
|
|
15
|
+
score: ok ? maxScore : 0,
|
|
16
|
+
reasoning: why
|
|
17
|
+
})
|
|
18
|
+
]
|
|
19
|
+
});
|
|
20
|
+
export const noPii = (maxScore = 2) => makeEvaluator((input) => {
|
|
21
|
+
const hits = piiPatterns.flatMap(([label, pattern]) => (pattern.test(input) ? [label] : []));
|
|
22
|
+
return Effect.succeed(one("no-pii", hits.length === 0, maxScore, hits.length === 0 ? "no PII detected" : `matched: ${hits.join(", ")}`));
|
|
23
|
+
});
|
|
24
|
+
export const validJson = (maxScore = 2) => makeEvaluator((input) => {
|
|
25
|
+
let valid = true;
|
|
26
|
+
try {
|
|
27
|
+
JSON.parse(input);
|
|
28
|
+
}
|
|
29
|
+
catch {
|
|
30
|
+
valid = false;
|
|
31
|
+
}
|
|
32
|
+
return Effect.succeed(one("valid-json", valid, maxScore, valid ? "parses as JSON" : "not valid JSON"));
|
|
33
|
+
});
|
|
34
|
+
export const matches = (regex, name = "format", maxScore = 2) => makeEvaluator((input) => {
|
|
35
|
+
const ok = new RegExp(`^(?:${regex})$`).test(input);
|
|
36
|
+
return Effect.succeed(one(name, ok, maxScore, ok ? `matches /${regex}/` : `does not match /${regex}/`));
|
|
37
|
+
});
|
|
38
|
+
export const lengthBetween = (min, max, maxScore = 2) => makeEvaluator((input) => Effect.succeed(one("length", input.length >= min && input.length <= max, maxScore, `length ${input.length}; bounds [${min}, ${max}]`)));
|
|
39
|
+
//# sourceMappingURL=Checks.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Checks.js","sourceRoot":"","sources":["../../src/eval/Checks.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,EAAE,cAAc,EAAE,UAAU,EAAE,MAAM,WAAW,CAAA;AACtD,OAAO,EAAE,aAAa,EAAkB,MAAM,gBAAgB,CAAA;AAE9D,MAAM,WAAW,GAA6D;IAC5E,CAAC,OAAO,EAAE,gDAAgD,CAAC;IAC3D,CAAC,KAAK,EAAE,uBAAuB,CAAC;IAChC,CAAC,aAAa,EAAE,wBAAwB,CAAC;IACzC,CAAC,OAAO,EAAE,mCAAmC,CAAC;IAC9C,CAAC,MAAM,EAAE,6BAA6B,CAAC;CACxC,CAAA;AAED,MAAM,GAAG,GAAG,CAAC,IAAY,EAAE,EAAW,EAAE,QAAgB,EAAE,GAAW,EAAc,EAAE,CACnF,UAAU,CAAC,IAAI,CAAC;IACd,MAAM,EAAE;QACN,cAAc,CAAC,IAAI,CAAC;YAClB,IAAI;YACJ,KAAK,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC;YACxB,SAAS,EAAE,GAAG;SACf,CAAC;KACH;CACF,CAAC,CAAA;AAEJ,MAAM,CAAC,MAAM,KAAK,GAAG,CAAC,QAAQ,GAAG,CAAC,EAAqB,EAAE,CACvD,aAAa,CAAC,CAAC,KAAK,EAAE,EAAE;IACtB,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAA;IAC5F,OAAO,MAAM,CAAC,OAAO,CACnB,GAAG,CACD,QAAQ,EACR,IAAI,CAAC,MAAM,KAAK,CAAC,EACjB,QAAQ,EACR,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,YAAY,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CACtE,CACF,CAAA;AACH,CAAC,CAAC,CAAA;AAEJ,MAAM,CAAC,MAAM,SAAS,GAAG,CAAC,QAAQ,GAAG,CAAC,EAAqB,EAAE,CAC3D,aAAa,CAAC,CAAC,KAAK,EAAE,EAAE;IACtB,IAAI,KAAK,GAAG,IAAI,CAAA;IAChB,IAAI,CAAC;QACH,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAA;IACnB,CAAC;IAAC,MAAM,CAAC;QACP,KAAK,GAAG,KAAK,CAAA;IACf,CAAC;IACD,OAAO,MAAM,CAAC,OAAO,CACnB,GAAG,CAAC,YAAY,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAChF,CAAA;AACH,CAAC,CAAC,CAAA;AAEJ,MAAM,CAAC,MAAM,OAAO,GAAG,CAAC,KAAa,EAAE,IAAI,GAAG,QAAQ,EAAE,QAAQ,GAAG,CAAC,EAAqB,EAAE,CACzF,aAAa,CAAC,CAAC,KAAK,EAAE,EAAE;IACtB,MAAM,EAAE,GAAG,IAAI,MAAM,CAAC,OAAO,KAAK,IAAI,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;IACnD,OAAO,MAAM,CAAC,OAAO,CACnB,GAAG,CAAC,IAAI,EAAE,EAAE,EAAE,QAAQ,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,GAAG,CAAC,CAAC,CAAC,mBAAmB,KAAK,GAAG,CAAC,CACjF,CAAA;AACH,CAAC,CAAC,CAAA;AAEJ,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,GAAW,EAAE,GAAW,EAAE,QAAQ,GAAG,CAAC,EAAqB,EAAE,CACzF,aAAa,CAAC,CAAC,KAAK,EAAE,EAAE,CACtB,MAAM,CAAC,OAAO,CACZ,GAAG,CACD,QAAQ,EACR,KAAK,CAAC,MAAM,IAAI,GAAG,IAAI,KAAK,CAAC,MAAM,IAAI,GAAG,EAC1C,QAAQ,EACR,UAAU,KAAK,CAAC,MAAM,aAAa,GAAG,KAAK,GAAG,GAAG,CAClD,CACF,CACF,CAAA"}
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import * as Schema from "effect/Schema";
|
|
3
|
+
import type { Evaluator } from "./Evaluator.ts";
|
|
4
|
+
declare const Dimension_base: Schema.Class<Dimension, Schema.Struct<{
|
|
5
|
+
readonly name: Schema.String;
|
|
6
|
+
readonly rubric: Schema.String;
|
|
7
|
+
readonly maxScore: Schema.withConstructorDefault<Schema.Int>;
|
|
8
|
+
}>, {}>;
|
|
9
|
+
export declare class Dimension extends Dimension_base {
|
|
10
|
+
}
|
|
11
|
+
declare const DimensionScore_base: Schema.Class<DimensionScore, Schema.Struct<{
|
|
12
|
+
readonly name: Schema.String;
|
|
13
|
+
readonly score: Schema.Int;
|
|
14
|
+
readonly reasoning: Schema.withConstructorDefault<Schema.String>;
|
|
15
|
+
}>, {}>;
|
|
16
|
+
export declare class DimensionScore extends DimensionScore_base {
|
|
17
|
+
}
|
|
18
|
+
declare const EvalResult_base: Schema.Class<EvalResult, Schema.Struct<{
|
|
19
|
+
readonly scores: Schema.$Array<typeof DimensionScore>;
|
|
20
|
+
readonly summary: Schema.withConstructorDefault<Schema.String>;
|
|
21
|
+
}>, {}>;
|
|
22
|
+
export declare class EvalResult extends EvalResult_base {
|
|
23
|
+
score(name: string): number | undefined;
|
|
24
|
+
get total(): number;
|
|
25
|
+
meets(minPerDimension: number): boolean;
|
|
26
|
+
}
|
|
27
|
+
declare const Sample_base: Schema.Class<Sample, Schema.Struct<{
|
|
28
|
+
readonly response: Schema.String;
|
|
29
|
+
readonly query: Schema.optionalKey<Schema.String>;
|
|
30
|
+
readonly context: Schema.optionalKey<Schema.String>;
|
|
31
|
+
readonly expected: Schema.optionalKey<Schema.String>;
|
|
32
|
+
}>, {}>;
|
|
33
|
+
export declare class Sample extends Sample_base {
|
|
34
|
+
}
|
|
35
|
+
declare const DimensionStats_base: Schema.Class<DimensionStats, Schema.Struct<{
|
|
36
|
+
readonly name: Schema.String;
|
|
37
|
+
readonly median: Schema.Int;
|
|
38
|
+
readonly min: Schema.Int;
|
|
39
|
+
readonly max: Schema.Int;
|
|
40
|
+
}>, {}>;
|
|
41
|
+
export declare class DimensionStats extends DimensionStats_base {
|
|
42
|
+
get spread(): number;
|
|
43
|
+
isFlaky(threshold: number): boolean;
|
|
44
|
+
}
|
|
45
|
+
declare const RepeatedEval_base: Schema.Class<RepeatedEval, Schema.Struct<{
|
|
46
|
+
readonly runs: Schema.$Array<typeof EvalResult>;
|
|
47
|
+
readonly stats: Schema.$Array<typeof DimensionStats>;
|
|
48
|
+
readonly spreadThreshold: Schema.Int;
|
|
49
|
+
}>, {}>;
|
|
50
|
+
export declare class RepeatedEval extends RepeatedEval_base {
|
|
51
|
+
get flakyDimensions(): ReadonlyArray<string>;
|
|
52
|
+
get isFlaky(): boolean;
|
|
53
|
+
get aggregate(): EvalResult;
|
|
54
|
+
}
|
|
55
|
+
export declare const repeat: <In>(evaluator: Evaluator<In>, input: In, n?: number | undefined, spreadThreshold?: number | undefined) => Effect.Effect<RepeatedEval, import("../Errors.ts").ProviderError | import("../Errors.ts").UsageLimitError | import("../Errors.ts").AuthenticationError | import("../Errors.ts").InvalidRequestError | import("../Errors.ts").ParseError | import("../Errors.ts").ToolError | import("../Errors.ts").ConfigError | import("../Errors.ts").RateLimitError | import("../Errors.ts").TimeoutError | import("../Errors.ts").TurnLimitError, never>;
|
|
56
|
+
export {};
|
|
57
|
+
//# sourceMappingURL=Eval.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Eval.d.ts","sourceRoot":"","sources":["../../src/eval/Eval.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAEvC,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAA;;;;;;AAE/C,qBAAa,SAAU,SAAQ,cAI7B;CAAG;;;;;;AAEL,qBAAa,cAAe,SAAQ,mBAIlC;CAAG;;;;;AAEL,qBAAa,UAAW,SAAQ,eAG9B;IACA,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS;IAIvC,IAAI,KAAK,IAAI,MAAM,CAElB;IAED,KAAK,CAAC,eAAe,EAAE,MAAM,GAAG,OAAO;CAGxC;;;;;;;AAED,qBAAa,MAAO,SAAQ,WAK1B;CAAG;;;;;;;AAEL,qBAAa,cAAe,SAAQ,mBAKlC;IACA,IAAI,MAAM,IAAI,MAAM,CAEnB;IAED,OAAO,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO;CAGpC;;;;;;AAED,qBAAa,YAAa,SAAQ,iBAIhC;IACA,IAAI,eAAe,IAAI,aAAa,CAAC,MAAM,CAAC,CAE3C;IAED,IAAI,OAAO,IAAI,OAAO,CAErB;IAED,IAAI,SAAS,IAAI,UAAU,CAW1B;CACF;AAKD,eAAO,MAAM,MAAM,GAAyD,EAAE,qhBAyB5E,CAAA"}
|
|
@@ -0,0 +1,91 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import * as Schema from "effect/Schema";
|
|
3
|
+
export class Dimension extends Schema.Class("Dimension")({
|
|
4
|
+
name: Schema.String,
|
|
5
|
+
rubric: Schema.String,
|
|
6
|
+
maxScore: Schema.Int.pipe(Schema.withConstructorDefault(Effect.succeed(2)))
|
|
7
|
+
}) {
|
|
8
|
+
}
|
|
9
|
+
export class DimensionScore extends Schema.Class("DimensionScore")({
|
|
10
|
+
name: Schema.String,
|
|
11
|
+
score: Schema.Int,
|
|
12
|
+
reasoning: Schema.String.pipe(Schema.withConstructorDefault(Effect.succeed("")))
|
|
13
|
+
}) {
|
|
14
|
+
}
|
|
15
|
+
export class EvalResult extends Schema.Class("EvalResult")({
|
|
16
|
+
scores: Schema.Array(DimensionScore),
|
|
17
|
+
summary: Schema.String.pipe(Schema.withConstructorDefault(Effect.succeed("")))
|
|
18
|
+
}) {
|
|
19
|
+
score(name) {
|
|
20
|
+
return this.scores.find((item) => item.name === name)?.score;
|
|
21
|
+
}
|
|
22
|
+
get total() {
|
|
23
|
+
return this.scores.reduce((sum, item) => sum + item.score, 0);
|
|
24
|
+
}
|
|
25
|
+
meets(minPerDimension) {
|
|
26
|
+
return this.scores.every((item) => item.score >= minPerDimension);
|
|
27
|
+
}
|
|
28
|
+
}
|
|
29
|
+
export class Sample extends Schema.Class("Sample")({
|
|
30
|
+
response: Schema.String,
|
|
31
|
+
query: Schema.optionalKey(Schema.String),
|
|
32
|
+
context: Schema.optionalKey(Schema.String),
|
|
33
|
+
expected: Schema.optionalKey(Schema.String)
|
|
34
|
+
}) {
|
|
35
|
+
}
|
|
36
|
+
export class DimensionStats extends Schema.Class("DimensionStats")({
|
|
37
|
+
name: Schema.String,
|
|
38
|
+
median: Schema.Int,
|
|
39
|
+
min: Schema.Int,
|
|
40
|
+
max: Schema.Int
|
|
41
|
+
}) {
|
|
42
|
+
get spread() {
|
|
43
|
+
return this.max - this.min;
|
|
44
|
+
}
|
|
45
|
+
isFlaky(threshold) {
|
|
46
|
+
return this.spread > threshold;
|
|
47
|
+
}
|
|
48
|
+
}
|
|
49
|
+
export class RepeatedEval extends Schema.Class("RepeatedEval")({
|
|
50
|
+
runs: Schema.Array(EvalResult),
|
|
51
|
+
stats: Schema.Array(DimensionStats),
|
|
52
|
+
spreadThreshold: Schema.Int
|
|
53
|
+
}) {
|
|
54
|
+
get flakyDimensions() {
|
|
55
|
+
return this.stats.filter((stat) => stat.isFlaky(this.spreadThreshold)).map((stat) => stat.name);
|
|
56
|
+
}
|
|
57
|
+
get isFlaky() {
|
|
58
|
+
return this.flakyDimensions.length > 0;
|
|
59
|
+
}
|
|
60
|
+
get aggregate() {
|
|
61
|
+
return EvalResult.make({
|
|
62
|
+
scores: this.stats.map((stat) => DimensionScore.make({
|
|
63
|
+
name: stat.name,
|
|
64
|
+
score: stat.median,
|
|
65
|
+
reasoning: `median of ${this.runs.length} runs`
|
|
66
|
+
})),
|
|
67
|
+
summary: `aggregate of ${this.runs.length} runs`
|
|
68
|
+
});
|
|
69
|
+
}
|
|
70
|
+
}
|
|
71
|
+
const median = (sorted) => sorted.length === 0 ? 0 : (sorted[Math.floor((sorted.length - 1) / 2)] ?? 0);
|
|
72
|
+
export const repeat = Effect.fn("@llm4ts/core/eval/Eval.repeat")(function* (evaluator, input, n = 3, spreadThreshold = 1) {
|
|
73
|
+
const runs = yield* Effect.forEach(Array.from({ length: Math.max(0, n) }), () => evaluator.evaluate(input));
|
|
74
|
+
const names = [...new Set(runs.flatMap((run) => run.scores.map((score) => score.name)))];
|
|
75
|
+
const stats = names.map((name) => {
|
|
76
|
+
const values = runs
|
|
77
|
+
.flatMap((run) => {
|
|
78
|
+
const score = run.score(name);
|
|
79
|
+
return score === undefined ? [] : [score];
|
|
80
|
+
})
|
|
81
|
+
.sort((left, right) => left - right);
|
|
82
|
+
return DimensionStats.make({
|
|
83
|
+
name,
|
|
84
|
+
median: median(values),
|
|
85
|
+
min: values.at(0) ?? 0,
|
|
86
|
+
max: values.at(-1) ?? 0
|
|
87
|
+
});
|
|
88
|
+
});
|
|
89
|
+
return RepeatedEval.make({ runs, stats, spreadThreshold });
|
|
90
|
+
});
|
|
91
|
+
//# sourceMappingURL=Eval.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Eval.js","sourceRoot":"","sources":["../../src/eval/Eval.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAIvC,MAAM,OAAO,SAAU,SAAQ,MAAM,CAAC,KAAK,CAAY,WAAW,CAAC,CAAC;IAClE,IAAI,EAAE,MAAM,CAAC,MAAM;IACnB,MAAM,EAAE,MAAM,CAAC,MAAM;IACrB,QAAQ,EAAE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,sBAAsB,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;CAC5E,CAAC;CAAG;AAEL,MAAM,OAAO,cAAe,SAAQ,MAAM,CAAC,KAAK,CAAiB,gBAAgB,CAAC,CAAC;IACjF,IAAI,EAAE,MAAM,CAAC,MAAM;IACnB,KAAK,EAAE,MAAM,CAAC,GAAG;IACjB,SAAS,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,sBAAsB,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC;CACjF,CAAC;CAAG;AAEL,MAAM,OAAO,UAAW,SAAQ,MAAM,CAAC,KAAK,CAAa,YAAY,CAAC,CAAC;IACrE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC;IACpC,OAAO,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,sBAAsB,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC;CAC/E,CAAC;IACA,KAAK,CAAC,IAAY;QAChB,OAAO,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,EAAE,KAAK,CAAA;IAC9D,CAAC;IAED,IAAI,KAAK;QACP,OAAO,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC,CAAA;IAC/D,CAAC;IAED,KAAK,CAAC,eAAuB;QAC3B,OAAO,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,IAAI,eAAe,CAAC,CAAA;IACnE,CAAC;CACF;AAED,MAAM,OAAO,MAAO,SAAQ,MAAM,CAAC,KAAK,CAAS,QAAQ,CAAC,CAAC;IACzD,QAAQ,EAAE,MAAM,CAAC,MAAM;IACvB,KAAK,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC;IACxC,OAAO,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC;IAC1C,QAAQ,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC;CAC5C,CAAC;CAAG;AAEL,MAAM,OAAO,cAAe,SAAQ,MAAM,CAAC,KAAK,CAAiB,gBAAgB,CAAC,CAAC;IACjF,IAAI,EAAE,MAAM,CAAC,MAAM;IACnB,MAAM,EAAE,MAAM,CAAC,GAAG;IAClB,GAAG,EAAE,MAAM,CAAC,GAAG;IACf,GAAG,EAAE,MAAM,CAAC,GAAG;CAChB,CAAC;IACA,IAAI,MAAM;QACR,OAAO,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAA;IAC5B,CAAC;IAED,OAAO,CAAC,SAAiB;QACvB,OAAO,IAAI,CAAC,MAAM,GAAG,SAAS,CAAA;IAChC,CAAC;CACF;AAED,MAAM,OAAO,YAAa,SAAQ,MAAM,CAAC,KAAK,CAAe,cAAc,CAAC,CAAC;IAC3E,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC;IAC9B,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC;IACnC,eAAe,EAAE,MAAM,CAAC,GAAG;CAC5B,CAAC;IACA,IAAI,eAAe;QACjB,OAAO,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IACjG,CAAC;IAED,IAAI,OAAO;QACT,OAAO,IAAI,CAAC,eAAe,CAAC,MAAM,GAAG,CAAC,CAAA;IACxC,CAAC;IAED,IAAI,SAAS;QACX,OAAO,UAAU,CAAC,IAAI,CAAC;YACrB,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAC9B,cAAc,CAAC,IAAI,CAAC;gBAClB,IAAI,EAAE,IAAI,CAAC,IAAI;gBACf,KAAK,EAAE,IAAI,CAAC,MAAM;gBAClB,SAAS,EAAE,aAAa,IAAI,CAAC,IAAI,CAAC,MAAM,OAAO;aAChD,CAAC,CACH;YACD,OAAO,EAAE,gBAAgB,IAAI,CAAC,IAAI,CAAC,MAAM,OAAO;SACjD,CAAC,CAAA;IACJ,CAAC;CACF;AAED,MAAM,MAAM,GAAG,CAAC,MAA6B,EAAU,EAAE,CACvD,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAA;AAE9E,MAAM,CAAC,MAAM,MAAM,GAAG,MAAM,CAAC,EAAE,CAAC,+BAA+B,CAAC,CAAC,QAAQ,CAAC,EACxE,SAAwB,EACxB,KAAS,EACT,CAAC,GAAG,CAAC,EACL,eAAe,GAAG,CAAC;IAEnB,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,GAAG,EAAE,CAC9E,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,CAC1B,CAAA;IACD,MAAM,KAAK,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAA;IACxF,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QAC/B,MAAM,MAAM,GAAG,IAAI;aAChB,OAAO,CAAC,CAAC,GAAG,EAAE,EAAE;YACf,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAA;YAC7B,OAAO,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAA;QAC3C,CAAC,CAAC;aACD,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,GAAG,KAAK,CAAC,CAAA;QACtC,OAAO,cAAc,CAAC,IAAI,CAAC;YACzB,IAAI;YACJ,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC;YACtB,GAAG,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;YACtB,GAAG,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;SACxB,CAAC,CAAA;IACJ,CAAC,CAAC,CAAA;IACF,OAAO,YAAY,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,eAAe,EAAE,CAAC,CAAA;AAC5D,CAAC,CAAC,CAAA"}
|
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import type { EvalResult, RepeatedEval } from "./Eval.ts";
|
|
3
|
+
import type { Evaluator } from "./Evaluator.ts";
|
|
4
|
+
export interface EvalCase<In> {
|
|
5
|
+
readonly name: string;
|
|
6
|
+
readonly input: In;
|
|
7
|
+
readonly minPerDimension?: number;
|
|
8
|
+
}
|
|
9
|
+
export interface CaseReport {
|
|
10
|
+
readonly name: string;
|
|
11
|
+
readonly result: EvalResult;
|
|
12
|
+
readonly repeated: RepeatedEval | undefined;
|
|
13
|
+
readonly passed: boolean;
|
|
14
|
+
}
|
|
15
|
+
export interface SuiteReport {
|
|
16
|
+
readonly cases: ReadonlyArray<CaseReport>;
|
|
17
|
+
readonly passed: boolean;
|
|
18
|
+
readonly failures: ReadonlyArray<CaseReport>;
|
|
19
|
+
}
|
|
20
|
+
export declare const run: <In>(evaluator: Evaluator<In>, cases: readonly EvalCase<In>[], repeats?: number | undefined, spreadThreshold?: number | undefined) => Effect.Effect<SuiteReport, import("../Errors.ts").ProviderError | import("../Errors.ts").UsageLimitError | import("../Errors.ts").AuthenticationError | import("../Errors.ts").InvalidRequestError | import("../Errors.ts").ParseError | import("../Errors.ts").ToolError | import("../Errors.ts").ConfigError | import("../Errors.ts").RateLimitError | import("../Errors.ts").TimeoutError | import("../Errors.ts").TurnLimitError, never>;
|
|
21
|
+
//# sourceMappingURL=EvalSuite.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"EvalSuite.d.ts","sourceRoot":"","sources":["../../src/eval/EvalSuite.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAGvC,OAAO,KAAK,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,WAAW,CAAA;AACzD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAA;AAE/C,MAAM,WAAW,QAAQ,CAAC,EAAE;IAC1B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;IACrB,QAAQ,CAAC,KAAK,EAAE,EAAE,CAAA;IAClB,QAAQ,CAAC,eAAe,CAAC,EAAE,MAAM,CAAA;CAClC;AAED,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;IACrB,QAAQ,CAAC,MAAM,EAAE,UAAU,CAAA;IAC3B,QAAQ,CAAC,QAAQ,EAAE,YAAY,GAAG,SAAS,CAAA;IAC3C,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAA;CACzB;AAED,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,KAAK,EAAE,aAAa,CAAC,UAAU,CAAC,CAAA;IACzC,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAA;IACxB,QAAQ,CAAC,QAAQ,EAAE,aAAa,CAAC,UAAU,CAAC,CAAA;CAC7C;AAED,eAAO,MAAM,GAAG,GAA2D,EAAE,+iBAqC3E,CAAA"}
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import { repeat } from "./Eval.js";
|
|
3
|
+
export const run = Effect.fn("@llm4ts/core/eval/EvalSuite.run")(function* (evaluator, cases, repeats = 1, spreadThreshold = 1) {
|
|
4
|
+
const reports = yield* Effect.forEach(cases, (testCase) => {
|
|
5
|
+
const minimum = testCase.minPerDimension ?? 1;
|
|
6
|
+
if (repeats <= 1) {
|
|
7
|
+
return evaluator.evaluate(testCase.input).pipe(Effect.map((result) => ({
|
|
8
|
+
name: testCase.name,
|
|
9
|
+
result,
|
|
10
|
+
repeated: undefined,
|
|
11
|
+
passed: result.meets(minimum)
|
|
12
|
+
})));
|
|
13
|
+
}
|
|
14
|
+
return repeat(evaluator, testCase.input, repeats, spreadThreshold).pipe(Effect.map((repeated) => {
|
|
15
|
+
const result = repeated.aggregate;
|
|
16
|
+
return {
|
|
17
|
+
name: testCase.name,
|
|
18
|
+
result,
|
|
19
|
+
repeated,
|
|
20
|
+
passed: result.meets(minimum) && !repeated.isFlaky
|
|
21
|
+
};
|
|
22
|
+
}));
|
|
23
|
+
});
|
|
24
|
+
return {
|
|
25
|
+
cases: reports,
|
|
26
|
+
passed: reports.every((report) => report.passed),
|
|
27
|
+
failures: reports.filter((report) => !report.passed)
|
|
28
|
+
};
|
|
29
|
+
});
|
|
30
|
+
//# sourceMappingURL=EvalSuite.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"EvalSuite.js","sourceRoot":"","sources":["../../src/eval/EvalSuite.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAEvC,OAAO,EAAE,MAAM,EAAE,MAAM,WAAW,CAAA;AAuBlC,MAAM,CAAC,MAAM,GAAG,GAAG,MAAM,CAAC,EAAE,CAAC,iCAAiC,CAAC,CAAC,QAAQ,CAAC,EACvE,SAAwB,EACxB,KAAkC,EAClC,OAAO,GAAG,CAAC,EACX,eAAe,GAAG,CAAC;IAEnB,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,EAAE;QACxD,MAAM,OAAO,GAAG,QAAQ,CAAC,eAAe,IAAI,CAAC,CAAA;QAC7C,IAAI,OAAO,IAAI,CAAC,EAAE,CAAC;YACjB,OAAO,SAAS,CAAC,QAAQ,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,IAAI,CAC5C,MAAM,CAAC,GAAG,CACR,CAAC,MAAM,EAAc,EAAE,CAAC,CAAC;gBACvB,IAAI,EAAE,QAAQ,CAAC,IAAI;gBACnB,MAAM;gBACN,QAAQ,EAAE,SAAS;gBACnB,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC;aAC9B,CAAC,CACH,CACF,CAAA;QACH,CAAC;QACD,OAAO,MAAM,CAAC,SAAS,EAAE,QAAQ,CAAC,KAAK,EAAE,OAAO,EAAE,eAAe,CAAC,CAAC,IAAI,CACrE,MAAM,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAc,EAAE;YAClC,MAAM,MAAM,GAAG,QAAQ,CAAC,SAAS,CAAA;YACjC,OAAO;gBACL,IAAI,EAAE,QAAQ,CAAC,IAAI;gBACnB,MAAM;gBACN,QAAQ;gBACR,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO;aACnD,CAAA;QACH,CAAC,CAAC,CACH,CAAA;IACH,CAAC,CAAC,CAAA;IACF,OAAO;QACL,KAAK,EAAE,OAAO;QACd,MAAM,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC;QAChD,QAAQ,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC;KACrD,CAAA;AACH,CAAC,CAAC,CAAA"}
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import type { LlmError } from "../Errors.ts";
|
|
3
|
+
import { EvalResult } from "./Eval.ts";
|
|
4
|
+
export interface Evaluator<in In> {
|
|
5
|
+
readonly evaluate: (input: In) => Effect.Effect<EvalResult, LlmError>;
|
|
6
|
+
}
|
|
7
|
+
export declare const makeEvaluator: <In>(evaluate: Evaluator<In>["evaluate"]) => Evaluator<In>;
|
|
8
|
+
export declare const contramap: <In, In2>(evaluator: Evaluator<In>, project: (input: In2) => In) => Evaluator<In2>;
|
|
9
|
+
export declare const all: <In>(...evaluators: ReadonlyArray<Evaluator<In>>) => Evaluator<In>;
|
|
10
|
+
//# sourceMappingURL=Evaluator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Evaluator.d.ts","sourceRoot":"","sources":["../../src/eval/Evaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,cAAc,CAAA;AAC5C,OAAO,EAAE,UAAU,EAAE,MAAM,WAAW,CAAA;AAEtC,MAAM,WAAW,SAAS,CAAC,EAAE,CAAC,EAAE;IAC9B,QAAQ,CAAC,QAAQ,EAAE,CAAC,KAAK,EAAE,EAAE,KAAK,MAAM,CAAC,MAAM,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;CACtE;AAED,eAAO,MAAM,aAAa,GAAI,EAAE,EAAE,UAAU,SAAS,CAAC,EAAE,CAAC,CAAC,UAAU,CAAC,KAAG,SAAS,CAAC,EAAE,CAElF,CAAA;AAEF,eAAO,MAAM,SAAS,GAAI,EAAE,EAAE,GAAG,EAC/B,WAAW,SAAS,CAAC,EAAE,CAAC,EACxB,SAAS,CAAC,KAAK,EAAE,GAAG,KAAK,EAAE,KAC1B,SAAS,CAAC,GAAG,CAAiE,CAAA;AAEjF,eAAO,MAAM,GAAG,GAAI,EAAE,EAAE,GAAG,YAAY,aAAa,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,KAAG,SAAS,CAAC,EAAE,CAa/E,CAAA"}
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import { EvalResult } from "./Eval.js";
|
|
3
|
+
export const makeEvaluator = (evaluate) => ({
|
|
4
|
+
evaluate
|
|
5
|
+
});
|
|
6
|
+
export const contramap = (evaluator, project) => makeEvaluator((input) => evaluator.evaluate(project(input)));
|
|
7
|
+
export const all = (...evaluators) => makeEvaluator((input) => Effect.forEach(evaluators, (evaluator) => evaluator.evaluate(input)).pipe(Effect.map((results) => EvalResult.make({
|
|
8
|
+
scores: results.flatMap((result) => result.scores),
|
|
9
|
+
summary: results
|
|
10
|
+
.map((result) => result.summary)
|
|
11
|
+
.filter((summary) => summary.length > 0)
|
|
12
|
+
.join("; ")
|
|
13
|
+
}))));
|
|
14
|
+
//# sourceMappingURL=Evaluator.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Evaluator.js","sourceRoot":"","sources":["../../src/eval/Evaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAEvC,OAAO,EAAE,UAAU,EAAE,MAAM,WAAW,CAAA;AAMtC,MAAM,CAAC,MAAM,aAAa,GAAG,CAAK,QAAmC,EAAiB,EAAE,CAAC,CAAC;IACxF,QAAQ;CACT,CAAC,CAAA;AAEF,MAAM,CAAC,MAAM,SAAS,GAAG,CACvB,SAAwB,EACxB,OAA2B,EACX,EAAE,CAAC,aAAa,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAA;AAEjF,MAAM,CAAC,MAAM,GAAG,GAAG,CAAK,GAAG,UAAwC,EAAiB,EAAE,CACpF,aAAa,CAAC,CAAC,KAAK,EAAE,EAAE,CACtB,MAAM,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CACvE,MAAM,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CACrB,UAAU,CAAC,IAAI,CAAC;IACd,MAAM,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC;IAClD,OAAO,EAAE,OAAO;SACb,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC;SAC/B,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC;SACvC,IAAI,CAAC,IAAI,CAAC;CACd,CAAC,CACH,CACF,CACF,CAAA"}
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
import * as Schema from "effect/Schema";
|
|
2
|
+
import type { LlmServiceShape } from "../LlmService.ts";
|
|
3
|
+
import { DimensionScore } from "./Eval.ts";
|
|
4
|
+
import type { Dimension, Sample } from "./Eval.ts";
|
|
5
|
+
import { type Evaluator } from "./Evaluator.ts";
|
|
6
|
+
declare const JudgeResponse_base: Schema.Class<JudgeResponse, Schema.Struct<{
|
|
7
|
+
readonly scores: Schema.$Array<typeof DimensionScore>;
|
|
8
|
+
}>, {}>;
|
|
9
|
+
export declare class JudgeResponse extends JudgeResponse_base {
|
|
10
|
+
}
|
|
11
|
+
export declare const defaultSystem: string;
|
|
12
|
+
export declare const buildPrompt: (system: string, dimensions: ReadonlyArray<Dimension>, sample: Sample) => string;
|
|
13
|
+
export declare const judge: (llm: LlmServiceShape, dimensions: ReadonlyArray<Dimension>, system?: string) => Evaluator<Sample>;
|
|
14
|
+
export {};
|
|
15
|
+
//# sourceMappingURL=Judge.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Judge.d.ts","sourceRoot":"","sources":["../../src/eval/Judge.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAA;AAEvD,OAAO,EAAE,cAAc,EAAc,MAAM,WAAW,CAAA;AACtD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,MAAM,WAAW,CAAA;AAClD,OAAO,EAAiB,KAAK,SAAS,EAAE,MAAM,gBAAgB,CAAA;;;;AAE9D,qBAAa,aAAc,SAAQ,kBAEjC;CAAG;AAEL,eAAO,MAAM,aAAa,QAEW,CAAA;AAsBrC,eAAO,MAAM,WAAW,GACtB,QAAQ,MAAM,EACd,YAAY,aAAa,CAAC,SAAS,CAAC,EACpC,QAAQ,MAAM,KACb,MAoBF,CAAA;AAsBD,eAAO,MAAM,KAAK,GAChB,KAAK,eAAe,EACpB,YAAY,aAAa,CAAC,SAAS,CAAC,EACpC,eAAsB,KACrB,SAAS,CAAC,MAAM,CAKhB,CAAA"}
|
|
@@ -0,0 +1,73 @@
|
|
|
1
|
+
import * as Effect from "effect/Effect";
|
|
2
|
+
import * as Schema from "effect/Schema";
|
|
3
|
+
import { DimensionScore, EvalResult } from "./Eval.js";
|
|
4
|
+
import { makeEvaluator } from "./Evaluator.js";
|
|
5
|
+
export class JudgeResponse extends Schema.Class("JudgeResponse")({
|
|
6
|
+
scores: Schema.Array(DimensionScore)
|
|
7
|
+
}) {
|
|
8
|
+
}
|
|
9
|
+
export const defaultSystem = "You are an impartial evaluator. Score each dimension strictly on its rubric; when unsure, score lower. " +
|
|
10
|
+
"Return ONLY valid JSON, no prose.";
|
|
11
|
+
const judgeJsonSchema = {
|
|
12
|
+
type: "object",
|
|
13
|
+
properties: {
|
|
14
|
+
scores: {
|
|
15
|
+
type: "array",
|
|
16
|
+
items: {
|
|
17
|
+
type: "object",
|
|
18
|
+
properties: {
|
|
19
|
+
name: { type: "string" },
|
|
20
|
+
score: { type: "integer" },
|
|
21
|
+
reasoning: { type: "string" }
|
|
22
|
+
},
|
|
23
|
+
required: ["name", "score"]
|
|
24
|
+
}
|
|
25
|
+
}
|
|
26
|
+
},
|
|
27
|
+
required: ["scores"],
|
|
28
|
+
additionalProperties: false
|
|
29
|
+
};
|
|
30
|
+
export const buildPrompt = (system, dimensions, sample) => {
|
|
31
|
+
const rubric = dimensions
|
|
32
|
+
.map((dimension) => `- ${dimension.name} (0..${dimension.maxScore}): ${dimension.rubric}`)
|
|
33
|
+
.join("\n");
|
|
34
|
+
const user = [
|
|
35
|
+
sample.query === undefined ? undefined : `Query: ${sample.query}`,
|
|
36
|
+
sample.context === undefined ? undefined : `Context: ${sample.context}`,
|
|
37
|
+
`Response: ${sample.response}`,
|
|
38
|
+
sample.expected === undefined ? undefined : `Expected: ${sample.expected}`
|
|
39
|
+
]
|
|
40
|
+
.filter((line) => line !== undefined)
|
|
41
|
+
.join("\n");
|
|
42
|
+
return `${system}
|
|
43
|
+
|
|
44
|
+
Score each dimension on its 0..max scale per the rubric:
|
|
45
|
+
${rubric}
|
|
46
|
+
|
|
47
|
+
Return ONLY this JSON: {"scores":[{"name":"<dimension>","score":<int>,"reasoning":"<one sentence>"}]}
|
|
48
|
+
|
|
49
|
+
${user}`;
|
|
50
|
+
};
|
|
51
|
+
const toResult = (dimensions, response) => {
|
|
52
|
+
const byName = new Map(response.scores.map((score) => [score.name, score]));
|
|
53
|
+
return EvalResult.make({
|
|
54
|
+
scores: dimensions.map((dimension) => {
|
|
55
|
+
const score = byName.get(dimension.name);
|
|
56
|
+
return score === undefined
|
|
57
|
+
? DimensionScore.make({
|
|
58
|
+
name: dimension.name,
|
|
59
|
+
score: 0,
|
|
60
|
+
reasoning: "missing"
|
|
61
|
+
})
|
|
62
|
+
: DimensionScore.make({
|
|
63
|
+
name: dimension.name,
|
|
64
|
+
score: Math.max(0, Math.min(dimension.maxScore, score.score)),
|
|
65
|
+
reasoning: score.reasoning
|
|
66
|
+
});
|
|
67
|
+
})
|
|
68
|
+
});
|
|
69
|
+
};
|
|
70
|
+
export const judge = (llm, dimensions, system = defaultSystem) => makeEvaluator((sample) => llm
|
|
71
|
+
.executeStructured(buildPrompt(system, dimensions, sample), JudgeResponse, judgeJsonSchema)
|
|
72
|
+
.pipe(Effect.map((response) => toResult(dimensions, response))));
|
|
73
|
+
//# sourceMappingURL=Judge.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Judge.js","sourceRoot":"","sources":["../../src/eval/Judge.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAGvC,OAAO,EAAE,cAAc,EAAE,UAAU,EAAE,MAAM,WAAW,CAAA;AAEtD,OAAO,EAAE,aAAa,EAAkB,MAAM,gBAAgB,CAAA;AAE9D,MAAM,OAAO,aAAc,SAAQ,MAAM,CAAC,KAAK,CAAgB,eAAe,CAAC,CAAC;IAC9E,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC;CACrC,CAAC;CAAG;AAEL,MAAM,CAAC,MAAM,aAAa,GACxB,yGAAyG;IACzG,mCAAmC,CAAA;AAErC,MAAM,eAAe,GAAe;IAClC,IAAI,EAAE,QAAQ;IACd,UAAU,EAAE;QACV,MAAM,EAAE;YACN,IAAI,EAAE,OAAO;YACb,KAAK,EAAE;gBACL,IAAI,EAAE,QAAQ;gBACd,UAAU,EAAE;oBACV,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBACxB,KAAK,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;oBAC1B,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;iBAC9B;gBACD,QAAQ,EAAE,CAAC,MAAM,EAAE,OAAO,CAAC;aAC5B;SACF;KACF;IACD,QAAQ,EAAE,CAAC,QAAQ,CAAC;IACpB,oBAAoB,EAAE,KAAK;CAC5B,CAAA;AAED,MAAM,CAAC,MAAM,WAAW,GAAG,CACzB,MAAc,EACd,UAAoC,EACpC,MAAc,EACN,EAAE;IACV,MAAM,MAAM,GAAG,UAAU;SACtB,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,KAAK,SAAS,CAAC,IAAI,QAAQ,SAAS,CAAC,QAAQ,MAAM,SAAS,CAAC,MAAM,EAAE,CAAC;SACzF,IAAI,CAAC,IAAI,CAAC,CAAA;IACb,MAAM,IAAI,GAAG;QACX,MAAM,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,MAAM,CAAC,KAAK,EAAE;QACjE,MAAM,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,MAAM,CAAC,OAAO,EAAE;QACvE,aAAa,MAAM,CAAC,QAAQ,EAAE;QAC9B,MAAM,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,aAAa,MAAM,CAAC,QAAQ,EAAE;KAC3E;SACE,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,IAAI,KAAK,SAAS,CAAC;SACpD,IAAI,CAAC,IAAI,CAAC,CAAA;IACb,OAAO,GAAG,MAAM;;;EAGhB,MAAM;;;;EAIN,IAAI,EAAE,CAAA;AACR,CAAC,CAAA;AAED,MAAM,QAAQ,GAAG,CAAC,UAAoC,EAAE,QAAuB,EAAc,EAAE;IAC7F,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,CAAA;IAC3E,OAAO,UAAU,CAAC,IAAI,CAAC;QACrB,MAAM,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE;YACnC,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,SAAS,CAAC,IAAI,CAAC,CAAA;YACxC,OAAO,KAAK,KAAK,SAAS;gBACxB,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC;oBAClB,IAAI,EAAE,SAAS,CAAC,IAAI;oBACpB,KAAK,EAAE,CAAC;oBACR,SAAS,EAAE,SAAS;iBACrB,CAAC;gBACJ,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC;oBAClB,IAAI,EAAE,SAAS,CAAC,IAAI;oBACpB,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;oBAC7D,SAAS,EAAE,KAAK,CAAC,SAAS;iBAC3B,CAAC,CAAA;QACR,CAAC,CAAC;KACH,CAAC,CAAA;AACJ,CAAC,CAAA;AAED,MAAM,CAAC,MAAM,KAAK,GAAG,CACnB,GAAoB,EACpB,UAAoC,EACpC,MAAM,GAAG,aAAa,EACH,EAAE,CACrB,aAAa,CAAC,CAAC,MAAM,EAAE,EAAE,CACvB,GAAG;KACA,iBAAiB,CAAC,WAAW,CAAC,MAAM,EAAE,UAAU,EAAE,MAAM,CAAC,EAAE,aAAa,EAAE,eAAe,CAAC;KAC1F,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC,CAAC,CAClE,CAAA"}
|
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
import * as Schema from "effect/Schema";
|
|
2
|
+
import type { LlmServiceShape } from "../LlmService.ts";
|
|
3
|
+
import type { MetricsCollectorShape, RequestLabels } from "./Metrics.ts";
|
|
4
|
+
declare const LlmPricing_base: Schema.Class<LlmPricing, Schema.Struct<{
|
|
5
|
+
readonly inputUsdPer1k: Schema.Number;
|
|
6
|
+
readonly outputUsdPer1k: Schema.Number;
|
|
7
|
+
}>, {}>;
|
|
8
|
+
export declare class LlmPricing extends LlmPricing_base {
|
|
9
|
+
}
|
|
10
|
+
export interface MeteredLlmOptions {
|
|
11
|
+
readonly labels: RequestLabels;
|
|
12
|
+
readonly pricing?: LlmPricing;
|
|
13
|
+
}
|
|
14
|
+
export declare const meterLlmService: (service: LlmServiceShape, collector: MetricsCollectorShape, options: MeteredLlmOptions) => LlmServiceShape;
|
|
15
|
+
export {};
|
|
16
|
+
//# sourceMappingURL=MeteredLlmService.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"MeteredLlmService.d.ts","sourceRoot":"","sources":["../../src/observability/MeteredLlmService.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAGvC,OAAO,KAAK,EAAE,eAAe,EAAoB,MAAM,kBAAkB,CAAA;AAGzE,OAAO,KAAK,EAAE,qBAAqB,EAAE,aAAa,EAAE,MAAM,cAAc,CAAA;;;;;AAExE,qBAAa,UAAW,SAAQ,eAG9B;CAAG;AAEL,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,MAAM,EAAE,aAAa,CAAA;IAC9B,QAAQ,CAAC,OAAO,CAAC,EAAE,UAAU,CAAA;CAC9B;AA+GD,eAAO,MAAM,eAAe,GAC1B,SAAS,eAAe,EACxB,WAAW,qBAAqB,EAChC,SAAS,iBAAiB,KACzB,eAqBD,CAAA"}
|
|
@@ -0,0 +1,66 @@
|
|
|
1
|
+
import * as Clock from "effect/Clock";
|
|
2
|
+
import * as Effect from "effect/Effect";
|
|
3
|
+
import * as Ref from "effect/Ref";
|
|
4
|
+
import * as Schema from "effect/Schema";
|
|
5
|
+
import * as Stream from "effect/Stream";
|
|
6
|
+
import { RequestMetrics } from "./Metrics.js";
|
|
7
|
+
export class LlmPricing extends Schema.Class("LlmPricing")({
|
|
8
|
+
inputUsdPer1k: Schema.Number,
|
|
9
|
+
outputUsdPer1k: Schema.Number
|
|
10
|
+
}) {
|
|
11
|
+
}
|
|
12
|
+
const estimatedCost = (usage, pricing) => usage === undefined || pricing === undefined
|
|
13
|
+
? undefined
|
|
14
|
+
: (usage.prompt / 1_000) * pricing.inputUsdPer1k +
|
|
15
|
+
(usage.completion / 1_000) * pricing.outputUsdPer1k;
|
|
16
|
+
const complete = (collector, options, startedAt, endedAt, success, errorType, usage) => {
|
|
17
|
+
const cost = estimatedCost(usage, options.pricing);
|
|
18
|
+
return collector.recordCompleted(RequestMetrics.make({
|
|
19
|
+
labels: options.labels,
|
|
20
|
+
latencyMs: Math.max(0, endedAt - startedAt),
|
|
21
|
+
success,
|
|
22
|
+
...(errorType === undefined ? {} : { errorType }),
|
|
23
|
+
...(usage === undefined ? {} : { tokenUsage: usage }),
|
|
24
|
+
...(cost === undefined ? {} : { estimatedCostUsd: cost })
|
|
25
|
+
}));
|
|
26
|
+
};
|
|
27
|
+
const meterEffect = (effect, collector, options, usageOf) => Effect.gen(function* () {
|
|
28
|
+
const startedAt = yield* Clock.currentTimeMillis;
|
|
29
|
+
yield* collector.markRequestStarted(options.labels);
|
|
30
|
+
const result = yield* Effect.result(effect);
|
|
31
|
+
const endedAt = yield* Clock.currentTimeMillis;
|
|
32
|
+
yield* complete(collector, options, startedAt, endedAt, result._tag === "Success", result._tag === "Failure" ? result.failure._tag : undefined, result._tag === "Success" ? usageOf(result.success) : undefined);
|
|
33
|
+
return yield* Effect.fromResult(result);
|
|
34
|
+
});
|
|
35
|
+
const meterStream = (stream, collector, options) => Stream.unwrap(Effect.gen(function* () {
|
|
36
|
+
const startedAt = yield* Clock.currentTimeMillis;
|
|
37
|
+
yield* collector.markRequestStarted(options.labels);
|
|
38
|
+
const measurement = yield* Ref.make({
|
|
39
|
+
success: true,
|
|
40
|
+
errorType: undefined,
|
|
41
|
+
usage: undefined
|
|
42
|
+
});
|
|
43
|
+
return stream.pipe(Stream.tap((chunk) => chunk.usage === undefined
|
|
44
|
+
? Effect.void
|
|
45
|
+
: Ref.update(measurement, (current) => ({
|
|
46
|
+
...current,
|
|
47
|
+
usage: chunk.usage
|
|
48
|
+
}))), Stream.tapError((error) => Ref.update(measurement, (current) => ({
|
|
49
|
+
...current,
|
|
50
|
+
success: false,
|
|
51
|
+
errorType: error._tag
|
|
52
|
+
}))), Stream.ensuring(Effect.gen(function* () {
|
|
53
|
+
const current = yield* Ref.get(measurement);
|
|
54
|
+
const endedAt = yield* Clock.currentTimeMillis;
|
|
55
|
+
yield* complete(collector, options, startedAt, endedAt, current.success, current.errorType, current.usage);
|
|
56
|
+
})));
|
|
57
|
+
}));
|
|
58
|
+
export const meterLlmService = (service, collector, options) => ({
|
|
59
|
+
executeStream: (prompt) => meterStream(service.executeStream(prompt), collector, options),
|
|
60
|
+
executeStreamWithHistory: (messages) => meterStream(service.executeStreamWithHistory(messages), collector, options),
|
|
61
|
+
executeWithTools: (prompt, tools) => meterEffect(service.executeWithTools(prompt, tools), collector, options, () => undefined),
|
|
62
|
+
executeStructured: (prompt, schema, jsonSchema) => meterEffect(service.executeStructured(prompt, schema, jsonSchema), collector, options, () => undefined),
|
|
63
|
+
executeStructuredWithUsage: (prompt, schema, jsonSchema) => meterEffect(service.executeStructuredWithUsage(prompt, schema, jsonSchema), collector, options, (result) => result[1]),
|
|
64
|
+
isAvailable: service.isAvailable
|
|
65
|
+
});
|
|
66
|
+
//# sourceMappingURL=MeteredLlmService.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"MeteredLlmService.js","sourceRoot":"","sources":["../../src/observability/MeteredLlmService.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,KAAK,MAAM,cAAc,CAAA;AACrC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,GAAG,MAAM,YAAY,CAAA;AACjC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AAIvC,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAA;AAG7C,MAAM,OAAO,UAAW,SAAQ,MAAM,CAAC,KAAK,CAAa,YAAY,CAAC,CAAC;IACrE,aAAa,EAAE,MAAM,CAAC,MAAM;IAC5B,cAAc,EAAE,MAAM,CAAC,MAAM;CAC9B,CAAC;CAAG;AAOL,MAAM,aAAa,GAAG,CACpB,KAA6B,EAC7B,OAA+B,EACX,EAAE,CACtB,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,SAAS;IAC1C,CAAC,CAAC,SAAS;IACX,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,GAAG,OAAO,CAAC,aAAa;QAC9C,CAAC,KAAK,CAAC,UAAU,GAAG,KAAK,CAAC,GAAG,OAAO,CAAC,cAAc,CAAA;AAEzD,MAAM,QAAQ,GAAG,CACf,SAAgC,EAChC,OAA0B,EAC1B,SAAiB,EACjB,OAAe,EACf,OAAgB,EAChB,SAA6B,EAC7B,KAA6B,EACR,EAAE;IACvB,MAAM,IAAI,GAAG,aAAa,CAAC,KAAK,EAAE,OAAO,CAAC,OAAO,CAAC,CAAA;IAClD,OAAO,SAAS,CAAC,eAAe,CAC9B,cAAc,CAAC,IAAI,CAAC;QAClB,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,SAAS,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;QAC3C,OAAO;QACP,GAAG,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC;QACjD,GAAG,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC;QACrD,GAAG,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,IAAI,EAAE,CAAC;KAC1D,CAAC,CACH,CAAA;AACH,CAAC,CAAA;AAED,MAAM,WAAW,GAAG,CAClB,MAAqC,EACrC,SAAgC,EAChC,OAA0B,EAC1B,OAA6C,EACd,EAAE,CACjC,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC;IAClB,MAAM,SAAS,GAAG,KAAK,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAA;IAChD,KAAK,CAAC,CAAC,SAAS,CAAC,kBAAkB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAA;IACnD,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAA;IAC3C,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAA;IAC9C,KAAK,CAAC,CAAC,QAAQ,CACb,SAAS,EACT,OAAO,EACP,SAAS,EACT,OAAO,EACP,MAAM,CAAC,IAAI,KAAK,SAAS,EACzB,MAAM,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EAC3D,MAAM,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAChE,CAAA;IACD,OAAO,KAAK,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,CAAA;AACzC,CAAC,CAAC,CAAA;AAQJ,MAAM,WAAW,GAAG,CAClB,MAAyC,EACzC,SAAgC,EAChC,OAA0B,EACS,EAAE,CACrC,MAAM,CAAC,MAAM,CACX,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC;IAClB,MAAM,SAAS,GAAG,KAAK,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAA;IAChD,KAAK,CAAC,CAAC,SAAS,CAAC,kBAAkB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAA;IACnD,MAAM,WAAW,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,IAAI,CAAoB;QACrD,OAAO,EAAE,IAAI;QACb,SAAS,EAAE,SAAS;QACpB,KAAK,EAAE,SAAS;KACjB,CAAC,CAAA;IACF,OAAO,MAAM,CAAC,IAAI,CAChB,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CACnB,KAAK,CAAC,KAAK,KAAK,SAAS;QACvB,CAAC,CAAC,MAAM,CAAC,IAAI;QACb,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACpC,GAAG,OAAO;YACV,KAAK,EAAE,KAAK,CAAC,KAAK;SACnB,CAAC,CAAC,CACR,EACD,MAAM,CAAC,QAAQ,CAAC,CAAC,KAAK,EAAE,EAAE,CACxB,GAAG,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACpC,GAAG,OAAO;QACV,OAAO,EAAE,KAAK;QACd,SAAS,EAAE,KAAK,CAAC,IAAI;KACtB,CAAC,CAAC,CACJ,EACD,MAAM,CAAC,QAAQ,CACb,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC;QAClB,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,WAAW,CAAC,CAAA;QAC3C,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAA;QAC9C,KAAK,CAAC,CAAC,QAAQ,CACb,SAAS,EACT,OAAO,EACP,SAAS,EACT,OAAO,EACP,OAAO,CAAC,OAAO,EACf,OAAO,CAAC,SAAS,EACjB,OAAO,CAAC,KAAK,CACd,CAAA;IACH,CAAC,CAAC,CACH,CACF,CAAA;AACH,CAAC,CAAC,CACH,CAAA;AAEH,MAAM,CAAC,MAAM,eAAe,GAAG,CAC7B,OAAwB,EACxB,SAAgC,EAChC,OAA0B,EACT,EAAE,CAAC,CAAC;IACrB,aAAa,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,WAAW,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,OAAO,CAAC;IACzF,wBAAwB,EAAE,CAAC,QAAQ,EAAE,EAAE,CACrC,WAAW,CAAC,OAAO,CAAC,wBAAwB,CAAC,QAAQ,CAAC,EAAE,SAAS,EAAE,OAAO,CAAC;IAC7E,gBAAgB,EAAE,CAAC,MAAM,EAAE,KAAK,EAAE,EAAE,CAClC,WAAW,CAAC,OAAO,CAAC,gBAAgB,CAAC,MAAM,EAAE,KAAK,CAAC,EAAE,SAAS,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,SAAS,CAAC;IAC3F,iBAAiB,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,EAAE,CAChD,WAAW,CACT,OAAO,CAAC,iBAAiB,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,CAAC,EACrD,SAAS,EACT,OAAO,EACP,GAAG,EAAE,CAAC,SAAS,CAChB;IACH,0BAA0B,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,EAAE,CACzD,WAAW,CACT,OAAO,CAAC,0BAA0B,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,CAAC,EAC9D,SAAS,EACT,OAAO,EACP,CAAI,MAA2B,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAC9C;IACH,WAAW,EAAE,OAAO,CAAC,WAAW;CACjC,CAAC,CAAA"}
|
|
@@ -0,0 +1,71 @@
|
|
|
1
|
+
import * as Context from "effect/Context";
|
|
2
|
+
import * as Effect from "effect/Effect";
|
|
3
|
+
import * as Layer from "effect/Layer";
|
|
4
|
+
import * as Schema from "effect/Schema";
|
|
5
|
+
import { TokenUsage } from "../Models.ts";
|
|
6
|
+
declare const RequestLabels_base: Schema.Class<RequestLabels, Schema.Struct<{
|
|
7
|
+
readonly provider: Schema.String;
|
|
8
|
+
readonly model: Schema.String;
|
|
9
|
+
readonly agent: Schema.optionalKey<Schema.String>;
|
|
10
|
+
readonly runId: Schema.optionalKey<Schema.String>;
|
|
11
|
+
readonly workflowStep: Schema.optionalKey<Schema.String>;
|
|
12
|
+
}>, {}>;
|
|
13
|
+
export declare class RequestLabels extends RequestLabels_base {
|
|
14
|
+
}
|
|
15
|
+
declare const RequestMetrics_base: Schema.Class<RequestMetrics, Schema.Struct<{
|
|
16
|
+
readonly labels: typeof RequestLabels;
|
|
17
|
+
readonly tokenUsage: Schema.optionalKey<typeof TokenUsage>;
|
|
18
|
+
readonly latencyMs: Schema.Number;
|
|
19
|
+
readonly success: Schema.Boolean;
|
|
20
|
+
readonly errorType: Schema.optionalKey<Schema.String>;
|
|
21
|
+
readonly estimatedCostUsd: Schema.optionalKey<Schema.Number>;
|
|
22
|
+
}>, {}>;
|
|
23
|
+
export declare class RequestMetrics extends RequestMetrics_base {
|
|
24
|
+
}
|
|
25
|
+
declare const ProviderHealth_base: Schema.Class<ProviderHealth, Schema.Struct<{
|
|
26
|
+
readonly provider: Schema.String;
|
|
27
|
+
readonly requestCount: Schema.Int;
|
|
28
|
+
readonly successRate: Schema.Number;
|
|
29
|
+
readonly avgLatencyMs: Schema.Number;
|
|
30
|
+
readonly p95LatencyMs: Schema.Number;
|
|
31
|
+
readonly estimatedCostUsd: Schema.Number;
|
|
32
|
+
}>, {}>;
|
|
33
|
+
export declare class ProviderHealth extends ProviderHealth_base {
|
|
34
|
+
}
|
|
35
|
+
declare const ObservabilitySnapshot_base: Schema.Class<ObservabilitySnapshot, Schema.Struct<{
|
|
36
|
+
readonly totalRequests: Schema.Int;
|
|
37
|
+
readonly totalErrors: Schema.Int;
|
|
38
|
+
readonly activeRequests: Schema.Int;
|
|
39
|
+
readonly promptTokens: Schema.Int;
|
|
40
|
+
readonly completionTokens: Schema.Int;
|
|
41
|
+
readonly totalTokens: Schema.Int;
|
|
42
|
+
readonly estimatedCostUsd: Schema.Number;
|
|
43
|
+
readonly p50LatencyMs: Schema.Number;
|
|
44
|
+
readonly p95LatencyMs: Schema.Number;
|
|
45
|
+
readonly byProvider: Schema.$Array<typeof ProviderHealth>;
|
|
46
|
+
readonly byAgentRequests: Schema.$Record<Schema.String, Schema.Int>;
|
|
47
|
+
}>, {}>;
|
|
48
|
+
export declare class ObservabilitySnapshot extends ObservabilitySnapshot_base {
|
|
49
|
+
}
|
|
50
|
+
declare const DashboardMetricsSnapshot_base: Schema.Class<DashboardMetricsSnapshot, Schema.Struct<{
|
|
51
|
+
readonly providerHealth: Schema.$Array<typeof ProviderHealth>;
|
|
52
|
+
readonly totalTokens: Schema.Int;
|
|
53
|
+
readonly totalCostUsd: Schema.Number;
|
|
54
|
+
readonly totalRequests: Schema.Int;
|
|
55
|
+
readonly errorRate: Schema.Number;
|
|
56
|
+
}>, {}>;
|
|
57
|
+
export declare class DashboardMetricsSnapshot extends DashboardMetricsSnapshot_base {
|
|
58
|
+
}
|
|
59
|
+
export interface MetricsCollectorShape {
|
|
60
|
+
readonly markRequestStarted: (labels: RequestLabels) => Effect.Effect<void>;
|
|
61
|
+
readonly recordCompleted: (metrics: RequestMetrics) => Effect.Effect<void>;
|
|
62
|
+
readonly snapshot: Effect.Effect<ObservabilitySnapshot>;
|
|
63
|
+
readonly dashboardSnapshot: Effect.Effect<DashboardMetricsSnapshot>;
|
|
64
|
+
}
|
|
65
|
+
declare const MetricsCollector_base: Context.ServiceClass<MetricsCollector, "@llm4ts/core/observability/MetricsCollector", MetricsCollectorShape>;
|
|
66
|
+
export declare class MetricsCollector extends MetricsCollector_base {
|
|
67
|
+
}
|
|
68
|
+
export declare const makeMetricsCollector: () => Effect.Effect<MetricsCollectorShape, never, never>;
|
|
69
|
+
export declare const MetricsCollectorLive: Layer.Layer<MetricsCollector, never, never>;
|
|
70
|
+
export {};
|
|
71
|
+
//# sourceMappingURL=Metrics.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"Metrics.d.ts","sourceRoot":"","sources":["../../src/observability/Metrics.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,OAAO,MAAM,gBAAgB,CAAA;AACzC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,KAAK,KAAK,MAAM,cAAc,CAAA;AAErC,OAAO,KAAK,MAAM,MAAM,eAAe,CAAA;AACvC,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAA;;;;;;;;AAEzC,qBAAa,aAAc,SAAQ,kBAMjC;CAAG;;;;;;;;;AAEL,qBAAa,cAAe,SAAQ,mBAOlC;CAAG;;;;;;;;;AAEL,qBAAa,cAAe,SAAQ,mBAOlC;CAAG;;;;;;;;;;;;;;AAEL,qBAAa,qBAAsB,SAAQ,0BAczC;CAAG;;;;;;;;AAEL,qBAAa,wBAAyB,SAAQ,6BAQ5C;CAAG;AAEL,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,kBAAkB,EAAE,CAAC,MAAM,EAAE,aAAa,KAAK,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAA;IAC3E,QAAQ,CAAC,eAAe,EAAE,CAAC,OAAO,EAAE,cAAc,KAAK,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAA;IAC1E,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,qBAAqB,CAAC,CAAA;IACvD,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC,MAAM,CAAC,wBAAwB,CAAC,CAAA;CACpE;;AAED,qBAAa,gBAAiB,SAAQ,qBAErC;CAAG;AA0EJ,eAAO,MAAM,oBAAoB,0DAkDhC,CAAA;AAED,eAAO,MAAM,oBAAoB,6CAAyD,CAAA"}
|