({ chat, apiKey, baseURL, model, messages, tools })
| 537 | } |
| 538 | |
| 539 | async function chatWithMinuteDelay({ chat, apiKey, baseURL, model, messages, tools }) { |
| 540 | const estimatedTokens = Math.max( |
| 541 | 1, |
| 542 | Math.ceil(JSON.stringify({ model, messages, ...(tools ? { tools } : {}) }).length / ESTIMATED_CHARS_PER_TOKEN) |
| 543 | ); |
| 544 | let currentMinuteBucket = Math.floor(Date.now() / 60000); |
| 545 | const estimateTokenUsage = (minuteTokenMap.get(currentMinuteBucket) ?? 0) + estimatedTokens; |
| 546 | |
| 547 | if (estimateTokenUsage > MAX_ESTIMATED_TOKENS_PER_MINUTE) { |
| 548 | currentMinuteBucket += 1; |
| 549 | await new Promise((resolve) => setTimeout(() => resolve(), 60000)); |
| 550 | } |
| 551 | |
| 552 | minuteTokenMap.set(currentMinuteBucket, (minuteTokenMap.get(currentMinuteBucket) ?? 0) + estimatedTokens); |
| 553 | |
| 554 | return { |
| 555 | minuteBucket: currentMinuteBucket, |
| 556 | response: await chat({ apiKey, baseURL, model, messages, tools }), |
| 557 | }; |
| 558 | } |
| 559 | |
| 560 | async function compactHistoryWithLLM(chat, apiKey, baseURL, model) { |
| 561 | const estimatedTokens = Math.max( |
no test coverage detected