Upload README.md with huggingface_hub

README.md

@@ -32,7 +32,7 @@ model never predicts the value directly).
 - live demo: https://huggingface.co/spaces/athrvk/gpu-sankhya-demo
 - gold evaluation sets: https://huggingface.co/datasets/athrvk/gpu-sankhya-gold
 
-This model card describes weights version `v0.3.3` (arch `v2`).
+This model card describes weights version `v0.3.4` (arch `v2`).
 
 ## Architecture
 
@@ -62,15 +62,15 @@ units, and range handling.
 | --- | ---: |
 | `charset.json` | 930 bytes |
 | `classes.json` | 1,571 bytes |
-| `gold_metrics_json_float32.json` | 4,314 bytes |
-| `gold_metrics_json_int8.json` | 4,295 bytes |
-| `gold_metrics_torch.json` | 4,314 bytes |
-| `kaggle_metrics.json` | 15,554 bytes |
-| `matrix.md` | 409 bytes |
+| `gold_metrics_json_float32.json` | 4,572 bytes |
+| `gold_metrics_json_int8.json` | 4,572 bytes |
+| `gold_metrics_torch.json` | 4,572 bytes |
+| `kaggle_metrics.json` | 16,439 bytes |
+| `matrix.md` | 739 bytes |
 | `sankhya.onnx` | 161,024 bytes |
-| `sankhya.pt` | 168,310 bytes |
-| `sankhya.weights.int8.json` | 58,997 bytes |
-| `sankhya.weights.json` | 397,100 bytes |
+| `sankhya.pt` | 167,990 bytes |
+| `sankhya.weights.int8.json` | 59,007 bytes |
+| `sankhya.weights.json` | 397,834 bytes |
 
 - `sankhya.pt` -- torch checkpoint (vocab, classes, arch, channels, state_dict)
 - `sankhya.onnx` -- ONNX graph, for interop/inspection
@@ -92,26 +92,26 @@ card](athrvk/gpu-sankhya-gold)):
 
 | gold set | examples | spans | value_acc | precision | recall | F1 |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: |
-| gold.jsonl | 220 | 194 | 0.9536 | 0.9585 | 0.9536 | 0.9561 |
-| gold_deva.jsonl | 182 | 151 | 0.9801 | 0.9867 | 0.9801 | 0.9834 |
-| **combined** | 402 | 345 | **0.9652** | 0.9708 | 0.9652 | 0.9680 |
+| gold.jsonl | 225 | 197 | 0.9543 | 0.9497 | 0.9594 | 0.9545 |
+| gold_deva.jsonl | 185 | 154 | 0.9740 | 0.9805 | 0.9805 | 0.9805 |
+| **combined** | 410 | 351 | **0.9630** | 0.9632 | 0.9687 | 0.9659 |
 
-Negatives (zero-gold-span examples): 73, false
+Negatives (zero-gold-span examples): 75, false
 positives: 0 (0.00%).
 
 Per-category value accuracy (combined):
 
 | category | spans | value_acc |
 | --- | ---: | ---: |
-| digits | 91 | 0.9780 |
-| words | 254 | 0.9606 |
+| digits | 92 | 0.9674 |
+| words | 259 | 0.9614 |
 | prefix | 81 | 0.9630 |
-| range | 24 | 0.9583 |
+| range | 27 | 0.8889 |
 | currency | 48 | 1.0000 |
-| multi_unit | 21 | 1.0000 |
+| multi_unit | 24 | 0.9167 |
 | symbol_unit | 33 | 1.0000 |
 | mixed_script | 2 | 1.0000 |
-| long | 10 | 0.8000 |
+| long | 12 | 0.6667 |
 
 ## Usage