Jasper Token Compression 600M — ONNX FP32
ONNX export of infgrad/Jasper-Token-Compression-600M.
Precision: FP32
Quantization: None (FP32)
Model size: 2.26 GiB
Full-precision FP32 ONNX export. This is the reference model used for the included quantization fidelity comparisons.
Benchmarks
| Tokens | Median latency | Tokens/s |
|---|---|---|
| 32 | 139.282 ms | 229.7 |
| 128 | 155.466 ms | 823.3 |
| 512 | 211.763 ms | 2,417.8 |
| 1024 | 288.388 ms | 3,550.8 |
Attribution
Original model: infgrad/Jasper-Token-Compression-600M