Update README.md

README.md

@@ -157,4 +157,63 @@ ckpt = torch.load(ckpt_path, map_location="cpu")
 
 print(ckpt.keys())          # e.g. ['config','epoch','encoder','projector','regularizer','predictor',...]
 print(ckpt["epoch"])        # e.g. 45
-encoder_state = ckpt["encoder"]
+encoder_state = ckpt["encoder"]
+
+---
+
+## `eval_runs/` contents (artifact index)
+
+This repo also includes full evaluation artifacts under `eval_runs/` for reproducibility.
+
+### CIFAR-100
+
+- `eval_runs/cifar100_hjepa_mv_80_epoch`
+- `eval_runs/cifar100_sigreg_tokens_80_epoch`
+- `eval_runs/cifar100_seed_42_30_epoch`
+- `eval_runs/cifar100_seed_45_30_epoch`
+- `eval_runs/cifar100_seed_49_30_epoch`
+
+Notes:
+- `*_80_epoch` folders are the long-run CIFAR-100 evaluations.
+- `cifar100_seed_{42,45,49}_30_epoch` are the 30-epoch seed runs used for mean±std reporting.
+
+### ImageNet-100 (main + checkpoint sweeps)
+
+Main 45-epoch evals:
+- `eval_runs/imagenet_hjepa_mv`
+- `eval_runs/imagenet_sigreg_tokens`
+
+Intermediate checkpoint evals:
+- `eval_runs/imagenet_hjepa_mv_epoch_010`
+- `eval_runs/imagenet_hjepa_mv_epoch_020`
+- `eval_runs/imagenet_hjepa_mv_epoch_030`
+- `eval_runs/imagenet_hjepa_mv_epoch_040`
+- `eval_runs/imagenet_sigreg_tokens_epoch_010`
+- `eval_runs/imagenet_sigreg_tokens_epoch_020`
+- `eval_runs/imagenet_sigreg_tokens_epoch_030`
+- `eval_runs/imagenet_sigreg_tokens_epoch_040`
+
+### ImageNet-100 linear-probe LR sweeps
+
+HamJEPA LR sweep:
+- `eval_runs/imagenet_hjepa_mv_lp_lr0p01`
+- `eval_runs/imagenet_hjepa_mv_lp_lr0p03`
+- `eval_runs/imagenet_hjepa_mv_lp_lr0p1`
+- `eval_runs/imagenet_hjepa_mv_lp_lr0p2`
+- `eval_runs/imagenet_hjepa_mv_lp_lr0p3`
+
+SIGReg LR sweep:
+- `eval_runs/imagenet_sigreg_tokens_lp_lr0p01`
+- `eval_runs/imagenet_sigreg_tokens_lp_lr0p03`
+- `eval_runs/imagenet_sigreg_tokens_lp_lr0p1`
+- `eval_runs/imagenet_sigreg_tokens_lp_lr0p2`
+- `eval_runs/imagenet_sigreg_tokens_lp_lr0p3`
+
+### Per-folder contents (typical)
+
+Each `eval_runs/...` folder contains:
+- `metrics.json` (primary machine-readable results)
+- plots in `plots/` (summary, kNN sweep, covariance/cosine/norm diagnostics)
+- auxiliary files produced by the evaluation scripts
+
+This structure is intended so all reported numbers can be traced directly to a corresponding `metrics.json`.