Sync updated model card and all charts

README.md

@@ -11,9 +11,9 @@ This repository contains checkpoints from the MPS-trained CIFAR-10 model
 comparison suite. Each `.pt` file preserves model weights, optimizer state,
 epoch history, architecture metadata, and validation-selection metadata.
 
-> **Fairness note:** DenseCNN was trained for 200 epochs; the other headline
-> models were trained for 30 epochs. DenseCNN's 92.53% is not a matched-budget
-> comparison. The best completed 30-epoch result is CNN-C + BN at 84.27%.
+> **Fairness note:** DenseCNN, CNN A/B/C, ViT, and MobileNet-A now have
+> 200-epoch results. R-CNN-B and MobileNet-B remain 30-epoch runs. Equal epoch
+> count does not imply identical optimization recipes or compute.
 
 ![CIFAR-10 test accuracy](accuracy_chart.svg)
 
@@ -42,13 +42,22 @@ inputs, batch size 128, 10 warmup iterations, and 50 measured iterations.
 | Model | Run epochs | Selected epoch | Best validation | Test accuracy |
 |---|---:|---:|---:|---:|
 | DenseCNN | 200* | **192** | 93.30% | **92.53%** |
-| CNN-C + BN | 30 | **27** | 86.24% | 84.27% |
-| CNN-C + LN | 30 | **30** | 84.00% | 82.74% |
+| CNN-C + BN | 200 | **176** | 89.70% | 88.48% |
+| CNN-C + LN | 200 | **123** | 88.90% | 87.73% |
+| CNN-C | 200 | **123** | 88.02% | 86.37% |
+| CNN-B + BN | 200 | **199** | 87.48% | 86.10% |
+| CNN-B + LN | 200 | **140** | 87.10% | 85.32% |
+| CNN-B | 200 | **148** | 86.36% | 84.43% |
+| CNN-A + LN | 200 | **185** | 82.04% | 81.51% |
+| CNN-A + BN | 200 | **179** | 82.18% | 81.23% |
 | R-CNN-B + BN | 30 | **29** | 82.40% | 81.22% |
-| MobileNet-A + BN | 30 | **29** | 62.24% | 62.68% |
-| MobileNet-A | 30 | **27** | 53.74% | 53.35% |
-| CNN-B + BN | 30 | **29** | 83.02% | 81.55% |
-| ViT | 30 | **28** | 70.66% | 69.08% |
+| CNN-A | 200 | **167** | 81.80% | 80.33% |
+| ViT | 200 | **182** | 80.02% | 79.31% |
+| MobileNet-A + BN | 200 | **198** | 70.20% | 68.78% |
+| MobileNet-B + BN | 30 | **29** | 66.22% | 66.28% |
+| MobileNet-A | 200 | **199** | 66.76% | 65.64% |
+| MobileNet-A + LN | 200 | **196** | 63.86% | 63.99% |
+| MobileNet-B | 30 | **30** | 56.92% | 57.02% |
 
 The full per-variant report is available in `reports/` in the source repository.
 
@@ -56,13 +65,14 @@ The full per-variant report is available in `reports/` in the source repository.
 checkpoint state is used for the reported test accuracy. **Run epochs** is the
 total training budget.
 
-`*` DenseCNN used a longer training budget; rerun all models for 200 epochs for
- a fair comparison.
+`*` DenseCNN uses a different optimization recipe from the baseline models;
+matched epoch count is not identical compute.
 
 | Checkpoint family | Notes |
 |---|---|
 | `cnn_A/B/C_{def,bn,ln}.pt` | CNN capacity and normalization variants |
 | `rcnn_B_bn.pt` | recurrent convolutional classifier |
+| `mobilenet_A/B_{def,bn,ln}.pt` | depthwise-separable MobileNet variants |
 | `vit.pt` | Vision Transformer |
 | `densecnn.pt` | DenseNet-inspired CNN; 92.53% test accuracy |
 | `dynamic_cnn.pt` | Dynamic capacity CNN |