Update README.md
Browse files
README.md
CHANGED
|
@@ -5,6 +5,63 @@ base_model:
|
|
| 5 |
pipeline_tag: zero-shot-classification
|
| 6 |
---
|
| 7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
# Updated - Spark works.
|
| 9 |
|
| 10 |
max-vit-goliath-spark is essentially a 300k param vit that can handle nearly identical accuracy as the larger model with a shockingly robust utility of the features.
|
|
|
|
| 5 |
pipeline_tag: zero-shot-classification
|
| 6 |
---
|
| 7 |
|
| 8 |
+
# Updated again - Spark has variants.
|
| 9 |
+
|
| 10 |
+
It works boys n grills. We have a micro-sized geometric ViT model that works.
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
```TEXT
|
| 14 |
+
Model Configuration:
|
| 15 |
+
Internal dim: 100
|
| 16 |
+
Vocab dim: 100
|
| 17 |
+
Num classes: 100
|
| 18 |
+
Crystal shape: torch.Size([100, 5, 100])
|
| 19 |
+
Evaluating: 100%|██████████| 100/100 [00:02<00:00, 37.96it/s]
|
| 20 |
+
|
| 21 |
+
================================================================================
|
| 22 |
+
EVALUATION RESULTS
|
| 23 |
+
================================================================================
|
| 24 |
+
|
| 25 |
+
Overall Accuracy: 53.50%
|
| 26 |
+
Auxiliary Head Accuracy: 52.97%
|
| 27 |
+
|
| 28 |
+
Top 10 Classes:
|
| 29 |
+
Class Acc% Conf GeoAlign CrystalNorm
|
| 30 |
+
----------------------------------------------------------------------
|
| 31 |
+
wardrobe 87.0 0.703 0.829 0.308
|
| 32 |
+
orange 84.0 0.708 0.839 0.298
|
| 33 |
+
road 84.0 0.772 0.626 0.327
|
| 34 |
+
sunflower 84.0 0.749 0.756 0.260
|
| 35 |
+
plain 80.0 0.692 0.763 0.306
|
| 36 |
+
skyscraper 80.0 0.669 0.631 0.255
|
| 37 |
+
apple 78.0 0.681 0.821 0.275
|
| 38 |
+
cloud 77.0 0.725 0.758 0.267
|
| 39 |
+
aquarium_fish 75.0 0.606 0.473 0.266
|
| 40 |
+
chair 73.0 0.709 0.696 0.279
|
| 41 |
+
|
| 42 |
+
Bottom 10 Classes:
|
| 43 |
+
Class Acc% Conf GeoAlign CrystalNorm
|
| 44 |
+
----------------------------------------------------------------------
|
| 45 |
+
kangaroo 33.0 0.434 0.601 0.316
|
| 46 |
+
man 33.0 0.461 0.554 0.321
|
| 47 |
+
squirrel 33.0 0.479 0.538 0.274
|
| 48 |
+
woman 33.0 0.399 0.576 0.289
|
| 49 |
+
boy 31.0 0.465 0.573 0.299
|
| 50 |
+
bus 31.0 0.526 0.694 0.298
|
| 51 |
+
possum 31.0 0.486 0.619 0.284
|
| 52 |
+
lizard 28.0 0.432 0.452 0.274
|
| 53 |
+
crocodile 25.0 0.408 0.481 0.310
|
| 54 |
+
seal 25.0 0.441 0.475 0.325
|
| 55 |
+
|
| 56 |
+
Correlations with Accuracy:
|
| 57 |
+
Geometric Alignment: 0.493
|
| 58 |
+
Crystal Norm: -0.210
|
| 59 |
+
Vertex Variance: -0.194
|
| 60 |
+
```
|
| 61 |
+

|
| 62 |
+
|
| 63 |
+
|
| 64 |
+
|
| 65 |
# Updated - Spark works.
|
| 66 |
|
| 67 |
max-vit-goliath-spark is essentially a 300k param vit that can handle nearly identical accuracy as the larger model with a shockingly robust utility of the features.
|