Output
ReadyGenerated text appears here. The prompt is shown muted and the continuation at full strength.
- Speed
- --tok/s
- First token
- --s
- Tokens
- --
- Model size
- --MB
Axiomic Labs' GDN2 and gated-GQA base model, exported to ONNX and running on CPU with ONNX Runtime. Pick a precision, write a prompt, and compare how each build streams.
Generated text appears here. The prompt is shown muted and the continuation at full strength.
Measured against the FP32 export: perplexity on a mixed text set, mean next-token KL divergence, and how often the top prediction matches FP32.
| Build | Size | Perplexity | KL vs FP32 | Top-1 agreement |
|---|