Cross-Framework Benchmarks
Every method tested across PyTorch, TensorFlow, and JAX. Same weights, same inputs, same results.
Threshold: Pearson r ≥ 0.95 for all framework pairs.
| Domain | Methods | Pass Rate | Min Correlation |
|---|---|---|---|
| Text (NLP) | 22 | 22/22 | r = 1.0000 |
| Image (CNN) | 22 | 22/22 | r = 0.9999 |
| Tabular (MLP) | 22 | 22/22 | r = 1.0000 |
| ECG (S4/Conv1D) | 22 | 22/22 | r = 0.9999 |
| Chess (Leela Zero) | 22 | 22/22 | r = 1.0000 |
| Protein (ESMFold) | 22 | 22/22 | r = 0.9999 |
| Audio (SSM) | 22 | 22/22 | r = 1.0000 |
| Graph (GNN) | 22 | 22/22 | r = 0.9999 |
| CLIP (Multimodal) | 18 | 18/18 | r = 0.9999 |
Results from automated validation pipeline. All tests run with identical model weights converted across frameworks.