Upload folder using huggingface_hub

README.md

@@ -12,6 +12,17 @@ size_categories:
   - 100K<n<1M
 task_categories:
   - image-classification
+dataset_info:
+  features:
+  - name: image
+    dtype: image
+  - name: label
+    dtype: int32
+  splits:
+  - name: train
+    num_examples: 503100
+  - name: validation
+    num_examples: 36758
 ---
 
 # Handwritten Digits Pack
@@ -31,13 +42,25 @@ Merged from 3 public datasets:
 
 ## Format
 
-- `X_train.pt.gz` — `torch.float32` tensor, shape `(503100, 1, 28, 28)`, range `[0.0, 1.0]`
-- `y_train.pt.gz` — `torch.int64` tensor, shape `(503100,)`, labels 0–9
-- `X_val.pt.gz` — `torch.float32` tensor, shape `(36758, 1, 28, 28)`
-- `y_val.pt.gz` — `torch.int64` tensor, shape `(36758,)`
+- `train.parquet` — 503,100 rows (191.8 MB)
+- `validation.parquet` — 36,758 rows (11.0 MB)
+- `image` column: PNG-encoded 28×28 grayscale (struct with `bytes` + `path` fields)
+- `label` column: int32, digits 0–9
+
+Also available as raw PyTorch tensors:
+- `X_train.pt.gz`, `y_train.pt.gz`, `X_val.pt.gz`, `y_val.pt.gz` — gzip-compressed float32 tensors
 
 ## Loading
 
+### Via `datasets` (recommended)
+
+```python
+from datasets import load_dataset
+ds = load_dataset("leobottaro/handwritten-digits-pack")
+```
+
+### Via raw tensors
+
 ```python
 import gzip, torch
 

train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a4918536d4ae6bf87b3d36fdb4a9dc323c7e92499097ee57a9dee0fdf05cb42
+size 201125233

validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:306e50ad74f30202f97207226d37db4b0a1ffe42a19e66b7cc501a516331b21e
+size 11492737