📊 TimesFM 2.5 200M Benchmark Report
Generated: 2026-06-29T13:05:32.950Z · Git: 32f6ac8e · Node: v22.23.0
· ONNX Runtime 1.27.0
💻 System
CPU
Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz × 4
🧠 Model
⚡ Inference Latency — Node.js vs WASM
| Context | Batch |
Node Avg (ms) | Node P99 (ms) |
WASM Avg (ms) | WASM P99 (ms) |
WASM/Node |
| 128 |
1 |
51.8 |
52.4 |
356.6 |
393.8 |
6.9× |
| 128 |
2 |
103.7 |
105.2 |
652.0 |
789.4 |
6.3× |
| 128 |
4 |
208.3 |
210.4 |
1283.4 |
1316.8 |
6.2× |
| 128 |
8 |
420.9 |
432.9 |
— |
— |
— |
| 256 |
1 |
52.9 |
54.2 |
329.9 |
334.5 |
6.2× |
| 256 |
2 |
105.4 |
107.0 |
657.8 |
659.3 |
6.2× |
| 256 |
4 |
208.1 |
209.6 |
1322.0 |
1352.7 |
6.4× |
| 256 |
8 |
417.7 |
421.8 |
— |
— |
— |
| 512 |
1 |
52.5 |
53.1 |
328.6 |
329.3 |
6.3× |
| 512 |
2 |
104.7 |
107.9 |
660.7 |
675.9 |
6.3× |
| 512 |
4 |
210.6 |
211.7 |
1320.6 |
1333.4 |
6.3× |
| 512 |
8 |
421.5 |
425.0 |
— |
— |
— |
🖥️ Node.js Inference Latency
| Context | Batch | Patches |
Avg (ms) | P50 (ms) | P99 (ms) |
Throughput (seq/s) | Cold Start (ms) |
| 128 |
1 |
4 |
51.8 |
51.9 |
52.4 |
19.3 |
53.2 |
| 128 |
2 |
4 |
103.7 |
104.1 |
105.2 |
9.6 |
51.6 |
| 128 |
4 |
4 |
208.3 |
208.3 |
210.4 |
4.8 |
51.2 |
| 128 |
8 |
4 |
420.9 |
421.7 |
432.9 |
2.4 |
51.8 |
| 256 |
1 |
8 |
52.9 |
53 |
54.2 |
18.9 |
52.3 |
| 256 |
2 |
8 |
105.4 |
105.3 |
107 |
9.5 |
52.8 |
| 256 |
4 |
8 |
208.1 |
207.8 |
209.6 |
4.8 |
52 |
| 256 |
8 |
8 |
417.7 |
417.9 |
421.8 |
2.4 |
52.1 |
| 512 |
1 |
16 |
52.5 |
52.5 |
53.1 |
19.1 |
53.6 |
| 512 |
2 |
16 |
104.7 |
104.5 |
107.9 |
9.5 |
52.9 |
| 512 |
4 |
16 |
210.6 |
210.7 |
211.7 |
4.7 |
52.8 |
| 512 |
8 |
16 |
421.5 |
422.4 |
425 |
2.4 |
52 |
🌐 WASM Inference Latency
| Context | Batch |
Avg (ms) | P50 (ms) | P99 (ms) |
Throughput (seq/s) | Cold Start (ms) |
| 128 |
1 |
356.6 |
336.8 |
393.8 |
2.8 |
366.8 |
| 128 |
2 |
652 |
628 |
789.4 |
1.5 |
391.8 |
| 128 |
4 |
1283.4 |
1283.7 |
1316.8 |
0.8 |
314 |
| 256 |
1 |
329.9 |
329.5 |
334.5 |
3 |
330.7 |
| 256 |
2 |
657.8 |
658.1 |
659.3 |
1.5 |
333.6 |
| 256 |
4 |
1322 |
1319.3 |
1352.7 |
0.8 |
330.6 |
| 512 |
1 |
328.6 |
328.6 |
329.3 |
3 |
329.3 |
| 512 |
2 |
660.7 |
657.2 |
675.9 |
1.5 |
329.2 |
| 512 |
4 |
1320.6 |
1317.7 |
1333.4 |
0.8 |
328.9 |
🎯 Prediction Accuracy
Full TimesFM pipeline with RevIN normalization, flip invariance, continuous quantile head
| Metric | Value |
|---|
| Naive MAE (no-change) | 87.7782 |
| Constant Mean MAE | 60.6676 |
| TimesFM MAE | 6.9042 |
| TimesFM RMSE | 8.4396 |
| Scaled MAE vs Naive | 0.0787 ✅ Better |
| Improvement vs Naive | 92.1% |
| Improvement vs Const Mean | 88.6% |
💾 Memory Footprint (Node.js)
| Metric | Value |
|---|
| RSS | 964 MB |
| Heap Used | 8.3 MB |
| Heap Total | 11.1 MB |