update model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# E2AM-ResNet50
|
| 2 |
+
|
| 3 |
+
## Model Details
|
| 4 |
+
- Architecture: ResNet-50
|
| 5 |
+
- Initialization: from scratch (`weights=None`)
|
| 6 |
+
- Dataset source: Kaggle ImageFolder or Hugging Face dataset; no torchvision-hosted dataset download is used.
|
| 7 |
+
- Energy: NVIDIA-SMI GPU power sampling summed across visible GPUs.
|
| 8 |
+
|
| 9 |
+
## Training Results
|
| 10 |
+
| run_id | variant_name | best_f1_score | final_accuracy | final_f1_score | total_energy_j | total_time_sec | total_co2_kg | num_parameters |
|
| 11 |
+
|:-----------------------|:---------------|----------------:|-----------------:|-----------------:|-----------------:|-----------------:|---------------:|-----------------:|
|
| 12 |
+
| cifar10/M2_amp_only | M2_amp_only | 0.666577 | 0.673 | 0.666577 | 477292 | 3803.68 | 0.062976 | 23528522 |
|
| 13 |
+
| cifar10/M7_full_e2am | M7_full_e2am | 0.683629 | 0.6851 | 0.683629 | 471771 | 3835.59 | 0.0622476 | 23528522 |
|
| 14 |
+
| cifar10/baseline_fixed | baseline_fixed | 0.661276 | 0.6546 | 0.661276 | 1.02776e+06 | 7988.1 | 0.135608 | 23528522 |
|
| 15 |
+
|
| 16 |
+
## Deployment Results
|
| 17 |
+
| variant | accuracy | f1_score | latency_ms_per_image | throughput_images_per_sec | energy_per_inference_j | sparsity_percent |
|
| 18 |
+
|:------------------------|-----------:|-----------:|-----------------------:|----------------------------:|-------------------------:|-------------------:|
|
| 19 |
+
| fp32 | 0.689063 | 0.687344 | 4.46075 | 224.178 | 0.435256 | 0 |
|
| 20 |
+
| pruned | 0.103906 | 0.0257757 | 3.33544 | 299.81 | 0.323063 | 19.9062 |
|
| 21 |
+
| pruned_finetuned | 0.665365 | 0.664818 | 3.18521 | 313.951 | 0.301914 | 10.7355 |
|
| 22 |
+
| dynamic_linear_int8_cpu | 0.686198 | 0.684309 | 225.033 | 4.4438 | nan | 0 |
|
| 23 |
+
|
| 24 |
+
## Limitations
|
| 25 |
+
- Dynamic depth routing is not claimed in this ResNet-50 experiment.
|
| 26 |
+
- CPU INT8 energy is not GPU energy and is reported as unavailable.
|