Upload README.md with huggingface_hub

README.md

@@ -27,16 +27,18 @@ Minimal JAX/Flax transformer for catalyst structure generation with single-param
 | Depth | 8 (n_embd=512, n_layer=8, n_head=8) |
 | Vocab size | 186 (WordLevel, 2-digit pair encoding) |
 | Training data | 174K OC20 structures |
-| Training time | 96 min on TPU v5p-8 |
+| Training time | 97 min on TPU v6e-8 |
 | Framework | JAX / Flax |
 
 ## Results (CuPt3 + OH, T=0.8, top_k=40, 100 samples)
 
 | Metric | Result |
 |--------|--------|
-| Generation Validity | 95/100 (95.0%) |
-| Uniqueness | 95/95 (100.0%) |
-| Novelty | 95/95 (100.0%) |
+| Parseable | 96/100 |
+| Element Match | 96/100 |
+| Generation Validity | 83/100 (83.0%) |
+| Uniqueness | 83/83 (100.0%) |
+| Novelty | 83/83 (100.0%) |
 
 ## Usage
 
@@ -54,7 +56,7 @@ config = CatalystConfig.load(ckpt_path / "config.json")
 
 ## Training
 
-Trained on 174K structures from the OC20 S2EF dataset using a TPU v5p-8 (Google TRC program). 20 epochs, WSD learning rate schedule, AdamW optimizer.
+Trained on 174K structures from the OC20 S2EF dataset using a TPU v6e-8 (Google TRC program). 20 epochs, WSD learning rate schedule, AdamW optimizer. Final val_loss=0.9518.
 
 ## Files