Add files using upload-large-folder tool

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+data/train_pairs.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,24 @@
+---
+pretty_name: SelectGround 8B Training Data
+configs:
+- config_name: default
+  data_files:
+  - split: train_pairs
+    path: data/train_pairs.jsonl
+  - split: train_replay
+    path: data/train_replay.jsonl
+  - split: refinement_pairs
+    path: data/refinement_pairs.jsonl
+  - split: refinement_replay
+    path: data/refinement_replay.jsonl
+---
+
+# SelectGround-8B training data
+
+This repository contains the exact data used to train SelectGround-8B. The main stage uses 3,790 mined target–distractor pairs and 3,790 replay examples. The final ten-step refinement uses 502 additional pairs and 502 matched replay examples.
+
+Each pair records an instruction, target point, target box, competing distractor box, available parser candidate boxes, and screenshot. Replay rows contain standard GUI grounding supervision. Coordinates and boxes use the original image size; responses are points normalized to `[0, 1000]`.
+
+The screenshots are a subset of [Click-100K](https://huggingface.co/datasets/mlfoundations/Click-100k). We redistribute only the examples required for reproducibility and retain the upstream dataset index for every row. Users remain responsible for the upstream data terms.
+
+See the [SelectGround repository](https://github.com/zhangruotian/SelectGround) for training code.

data/train_pairs.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07bd539b169a13fca1a9c7386869b5f865e6dbcc6fe9b546663903254b4afe6c
+size 43988851

dataset_info.json

@@ -0,0 +1,9 @@
+{
+  "splits": {
+    "train_pairs": 3790,
+    "train_replay": 3790,
+    "refinement_pairs": 502,
+    "refinement_replay": 502
+  },
+  "images": 6690
+}