| import type { |
| Api, |
| Context, |
| Model, |
| SimpleStreamOptions, |
| StreamOptions, |
| ThinkingBudgets, |
| ThinkingLevel, |
| } from "../types.ts"; |
| import { estimateContextTokens } from "../utils/estimate.ts"; |
|
|
| const CONTEXT_SAFETY_TOKENS = 4096; |
| const MIN_MAX_TOKENS = 1; |
|
|
| export function clampMaxTokensToContext(model: Model<Api>, context: Context, maxTokens: number): number { |
| if (model.contextWindow <= 0) return Math.max(MIN_MAX_TOKENS, maxTokens); |
| const available = model.contextWindow - estimateContextTokens(context).tokens - CONTEXT_SAFETY_TOKENS; |
| return Math.min(maxTokens, Math.max(MIN_MAX_TOKENS, available)); |
| } |
|
|
| export function buildBaseOptions( |
| model: Model<Api>, |
| context: Context, |
| options?: SimpleStreamOptions, |
| apiKey?: string, |
| ): StreamOptions { |
| const samplingParams = |
| model.samplingParams || options?.samplingParams |
| ? { ...model.samplingParams, ...options?.samplingParams } |
| : undefined; |
| return { |
| temperature: options?.temperature, |
| samplingParams, |
| maxTokens: clampMaxTokensToContext(model, context, options?.maxTokens ?? model.maxTokens), |
| signal: options?.signal, |
| telemetryContext: options?.telemetryContext, |
| apiKey: apiKey || options?.apiKey, |
| fetch: options?.fetch, |
| transport: options?.transport, |
| cacheRetention: options?.cacheRetention, |
| sessionId: options?.sessionId, |
| headers: options?.headers, |
| onPayload: options?.onPayload, |
| onResponse: options?.onResponse, |
| timeoutMs: options?.timeoutMs, |
| websocketConnectTimeoutMs: options?.websocketConnectTimeoutMs, |
| maxRetries: options?.maxRetries, |
| maxRetryDelayMs: options?.maxRetryDelayMs, |
| metadata: options?.metadata, |
| env: options?.env, |
| }; |
| } |
|
|
| |
| export const MIN_ANSWER_TOKENS = 1024; |
|
|
| export const DEFAULT_THINKING_BUDGETS: ThinkingBudgets = { |
| minimal: 1024, |
| low: 2048, |
| medium: 8192, |
| high: 16384, |
| }; |
|
|
| export function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, "xhigh" | "max"> | undefined { |
| return effort === "xhigh" || effort === "max" ? "high" : effort; |
| } |
|
|
| export function thinkingBudgetForLevel(reasoningLevel: ThinkingLevel, customBudgets?: ThinkingBudgets): number { |
| const budgets = { ...DEFAULT_THINKING_BUDGETS, ...customBudgets }; |
| const level = clampReasoning(reasoningLevel)!; |
| return budgets[level]!; |
| } |
|
|
| |
| export function clampThinkingBudgetToAnswerRoom(thinkingBudget: number, ceiling: number): number { |
| return Math.min(thinkingBudget, Math.max(0, ceiling - MIN_ANSWER_TOKENS)); |
| } |
|
|
| export function adjustMaxTokensForThinking( |
| |
| baseMaxTokens: number | undefined, |
| modelMaxTokens: number, |
| reasoningLevel: ThinkingLevel, |
| customBudgets?: ThinkingBudgets, |
| ): { maxTokens: number; thinkingBudget: number } { |
| let thinkingBudget = thinkingBudgetForLevel(reasoningLevel, customBudgets); |
| const maxTokens = |
| baseMaxTokens === undefined ? modelMaxTokens : Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens); |
|
|
| if (maxTokens <= thinkingBudget) { |
| thinkingBudget = clampThinkingBudgetToAnswerRoom(thinkingBudget, maxTokens); |
| } |
|
|
| return { maxTokens, thinkingBudget }; |
| } |
|
|
