{"version":3,"file":"simple-options.d.ts","sourceRoot":"","sources":["../../src/api/simple-options.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,GAAG,EACH,OAAO,EACP,KAAK,EACL,mBAAmB,EACnB,aAAa,EACb,eAAe,EACf,aAAa,EACb,MAAM,aAAa,CAAC;AAMrB,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAItG;AAED,wBAAgB,gBAAgB,CAC/B,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,EAC7B,MAAM,CAAC,EAAE,MAAM,GACb,aAAa,CA0Bf;AAED,4FAA4F;AAC5F,eAAO,MAAM,iBAAiB,OAAO,CAAC;AAEtC,wBAAgB,cAAc,CAAC,MAAM,EAAE,aAAa,GAAG,SAAS,GAAG,OAAO,CAAC,aAAa,EAAE,OAAO,GAAG,KAAK,CAAC,GAAG,SAAS,CAErH;AAED,wBAAgB,0BAA0B,CAEzC,aAAa,EAAE,MAAM,GAAG,SAAS,EACjC,cAAc,EAAE,MAAM,EACtB,cAAc,EAAE,aAAa,EAC7B,aAAa,CAAC,EAAE,eAAe,GAC7B;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,MAAM,CAAA;CAAE,CAmB/C","sourcesContent":["import type {\n\tApi,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamOptions,\n\tThinkingBudgets,\n\tThinkingLevel,\n} from \"../types.ts\";\nimport { estimateContextTokens } from \"../utils/estimate.ts\";\n\nconst CONTEXT_SAFETY_TOKENS = 4096;\nconst MIN_MAX_TOKENS = 1;\n\nexport function clampMaxTokensToContext(model: Model<Api>, context: Context, maxTokens: number): number {\n\tif (model.contextWindow <= 0) return Math.max(MIN_MAX_TOKENS, maxTokens);\n\tconst available = model.contextWindow - estimateContextTokens(context).tokens - CONTEXT_SAFETY_TOKENS;\n\treturn Math.min(maxTokens, Math.max(MIN_MAX_TOKENS, available));\n}\n\nexport function buildBaseOptions(\n\tmodel: Model<Api>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n\tapiKey?: string,\n): StreamOptions {\n\tconst samplingParams =\n\t\tmodel.samplingParams || options?.samplingParams\n\t\t\t? { ...model.samplingParams, ...options?.samplingParams }\n\t\t\t: undefined;\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tsamplingParams,\n\t\tmaxTokens: clampMaxTokensToContext(model, context, options?.maxTokens ?? model.maxTokens),\n\t\tsignal: options?.signal,\n\t\ttelemetryContext: options?.telemetryContext,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tfetch: options?.fetch,\n\t\ttransport: options?.transport,\n\t\tcacheRetention: options?.cacheRetention,\n\t\tsessionId: options?.sessionId,\n\t\theaders: options?.headers,\n\t\tonPayload: options?.onPayload,\n\t\tonResponse: options?.onResponse,\n\t\ttimeoutMs: options?.timeoutMs,\n\t\twebsocketConnectTimeoutMs: options?.websocketConnectTimeoutMs,\n\t\tmaxRetries: options?.maxRetries,\n\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\tmetadata: options?.metadata,\n\t\tenv: options?.env,\n\t};\n}\n\n/** Tokens always left for the answer when a thinking budget shares the response ceiling. */\nexport const MIN_ANSWER_TOKENS = 1024;\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\" | \"max\"> | undefined {\n\treturn effort === \"xhigh\" || effort === \"max\" ? \"high\" : effort;\n}\n\nexport function adjustMaxTokensForThinking(\n\t// Undefined means no explicit caller cap. Use the model cap and fit thinking inside it.\n\tbaseMaxTokens: number | undefined,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tconst defaultBudgets: ThinkingBudgets = {\n\t\tminimal: 1024,\n\t\tlow: 2048,\n\t\tmedium: 8192,\n\t\thigh: 16384,\n\t};\n\tconst budgets = { ...defaultBudgets, ...customBudgets };\n\n\tconst level = clampReasoning(reasoningLevel)!;\n\tlet thinkingBudget = budgets[level]!;\n\tconst maxTokens =\n\t\tbaseMaxTokens === undefined ? modelMaxTokens : Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = Math.max(0, maxTokens - MIN_ANSWER_TOKENS);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}