| { | |
| "data": { | |
| "dataset": "scene_parse_150 (ADE20K, 150 classes)", | |
| "image_size": 224, | |
| "patch_size": 16, | |
| "num_classes": 151, | |
| "ignore_index": 0 | |
| }, | |
| "model": { | |
| "architecture": "ViT-Segmentation (first-principles backbone + linear decoder head)", | |
| "embed_dim": 384, | |
| "depth": 8, | |
| "num_heads": 6, | |
| "mlp_ratio": 4.0, | |
| "dropout": 0.1 | |
| }, | |
| "training": { | |
| "batch_size": 8, | |
| "epochs": 5, | |
| "lr": 0.0003, | |
| "weight_decay": 0.05, | |
| "warmup_epochs": 1, | |
| "num_workers": 2, | |
| "subset_size": null, | |
| "device": "cuda if available else cpu" | |
| }, | |
| "evolution_search": { | |
| "population_size": 6, | |
| "generations": 4, | |
| "elite_k": 2, | |
| "steps_per_candidate": 30, | |
| "subset_size": 64, | |
| "batch_size": 4 | |
| } | |
| } |