Add DiT-XL/2 head (final epoch, from the original attention-mim run); 37/37 encoders covered

README.md

@@ -10,17 +10,19 @@ library_name: pytorch
 
 # Efficient Probing heads
 
-Trained EP (efficient probing) heads for 36 frozen encoders, from the standing
+Trained EP (efficient probing) heads for all 37 frozen encoders of the standing
 ImageNet-1k benchmark at [billpsomas/efficient-probing](https://github.com/billpsomas/efficient-probing)
 (ICLR 2026). Each head is a multi-query cross-attention pooling + BatchNorm + linear
 classifier, trained for 90 epochs with LARS on frozen features. **No backbone weights
 are included** -- each `config.json` records the exact `--model` / loader flags to
 rebuild the encoder from its original source.
 
-Every head ships with its accuracy **at the saved epoch** in `meta`/`config.json`:
-12 heads are the run's peak (they reproduce the leaderboard number exactly); the rest
-are final-epoch heads, with both figures recorded so nothing is ambiguous. DiT-XL/2
-is absent (no surviving checkpoint; see the manifest).
+Why two kinds of head: the training code initially kept only a rolling per-epoch
+checkpoint, so a finished run left its **final** epoch behind rather than its best.
+12 early-peaking models were later re-run with best-epoch checkpointing, so their
+heads are the **peak** and reproduce the leaderboard number exactly. Every file's
+metadata records both its own accuracy at the saved epoch and the table's
+best-epoch figure, so nothing has to be taken on trust.
 
 ## Loading
 
@@ -85,6 +87,7 @@ python tools/eval_reimagenet.py predict <backbone flags from config.json> \
 | MaskFeat ViT-L/16 | ep | 69.56 | 89 | final epoch |
 | SimMIM ViT-B/16 | ep | 64.81 | 89 | final epoch |
 | MAE ViT-S/16 | ep | 64.56 | 89 | final epoch |
+| DiT DiT-XL/2 | ep | 56.94 | 86 | final epoch |
 
 Full provenance (training logs, exact commands, the leaderboard itself) lives in the
 [GitHub repo](https://github.com/billpsomas/efficient-probing). Heads were trained on

dit_xl/config.json

@@ -0,0 +1,19 @@
+{
+ "method": "DiT",
+ "arch": "DiT-XL/2",
+ "pretrain": "IN-1K",
+ "image_size": 256,
+ "model": "DiT-XL/2",
+ "loader": "dit",
+ "pretrain_tag": "",
+ "finetune": "",
+ "extra_flags": "",
+ "cls_features": "ep",
+ "ep_queries": 128,
+ "d_out": 1,
+ "head_epoch": 86,
+ "val_acc1_at_head_epoch": 56.94,
+ "table_best_acc1": 57.0,
+ "note": "final-epoch head, not best-epoch: checkpoints overwrite per epoch. table_best_acc1 is the peak of the same run; the difference is the post-peak decline of a frozen-feature probe, largest on VLM encoders.",
+ "training_log": "logs/dit_xl/ep.txt"
+}

dit_xl/ep_head.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c641c72c0ddd6d090c605aaf6d5237a7c0ed46509140c30e2f365ab0747fc2d8
+size 10523089

manifest.json

@@ -330,6 +330,28 @@
    "file": "siglip2_vitl/ep_head.pth",
    "size_mb": 8.4
   },
+  {
+   "method": "DiT",
+   "arch": "DiT-XL/2",
+   "pretrain": "IN-1K",
+   "image_size": 256,
+   "model": "DiT-XL/2",
+   "loader": "dit",
+   "pretrain_tag": "",
+   "finetune": "",
+   "extra_flags": "",
+   "cls_features": "ep",
+   "ep_queries": 128,
+   "d_out": 1,
+   "head_epoch": 86,
+   "val_acc1_at_head_epoch": 56.94,
+   "table_best_acc1": 57.0,
+   "note": "final-epoch head, not best-epoch: checkpoints overwrite per epoch. table_best_acc1 is the peak of the same run; the difference is the post-peak decline of a frozen-feature probe, largest on VLM encoders.",
+   "training_log": "logs/dit_xl/ep.txt",
+   "params_incl_bn_stats": 2629865,
+   "file": "dit_xl/ep_head.pth",
+   "size_mb": 10.5
+  },
   {
    "method": "AIMv2",
    "arch": "ViT-L/14",
@@ -793,11 +815,5 @@
    "size_mb": 8.4
   }
  ],
- "missing": [
-  {
-   "method": "DiT",
-   "arch": "DiT-XL/2",
-   "reason": "no surviving checkpoint: the run's output directory was deleted after its logs were published; re-run required to export a head"
-  }
- ]
+ "missing": []
 }