Function suggestBatchSize

Suggest a perCoreBatchSize based on available system memory.

TimesFM 2.5 200M loads ~1.5 GB into RAM (model weights + activations). Each additional batch element adds ~200 MB for intermediate tensors. This function computes a safe batch size that fits within a configurable fraction of free memory (default: 50 %).

The result is clamped to [1, 16] — the model processes elements concurrently via Promise.all, so diminishing returns apply beyond 8.

import { suggestBatchSize, createForecastConfig } from '@agentix-e/timesfm-core';

const bs = suggestBatchSize();
// or with explicit options:
const bs = suggestBatchSize({ freeMemoryGB: 8, memoryFraction: 0.5 });
const fc = createForecastConfig({ perCoreBatchSize: bs });
Copy