From f309be2a172572be4c208dfbac3deba03c9e30eb Mon Sep 17 00:00:00 2001
From: pctablet505 <pctablet505@gmail.com>
Date: Wed, 15 Jul 2026 16:45:26 +0000
Subject: [PATCH] docs: add WebDataset example for multiple media files of the
 same type

Adds a WebDataset example to the video dataset docs showing how multiple
media files of the same type per sample are mapped to separate columns.

Fixes #7413
---
 docs/source/video_dataset.mdx | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/docs/source/video_dataset.mdx b/docs/source/video_dataset.mdx
index cb0c70485e4..8c1323bb96a 100644
--- a/docs/source/video_dataset.mdx
+++ b/docs/source/video_dataset.mdx
@@ -171,6 +171,30 @@ Load your WebDataset and it will create on column per file suffix (here "mp4" an
 {"bbox": [[302.0, 109.0, 73.0, 52.0]], "categories": [0]}
 ```
 
+If a sample contains several media files of the same type, use a period-separated name before the extension. For example, in a sign-language dataset you may have a base video and a separate pose video:
+
+```
+e39871fd9fd74f55.base.mp4
+e39871fd9fd74f55.pose.mp4
+e39871fd9fd74f55.json
+f18b91585c4d3f3e.base.mp4
+f18b91585c4d3f3e.pose.mp4
+f18b91585c4d3f3e.json
+...
+```
+
+The files are still grouped by the shared prefix, and each file becomes its own column. In this example, the dataset has the columns `base.mp4`, `pose.mp4` and `json`:
+
+```python
+>>> from datasets import load_dataset
+
+>>> dataset = load_dataset("webdataset", data_dir="/path/to/folder", split="train")
+>>> sorted(dataset[0].keys())
+['__key__', '__url__', 'base.mp4', 'json', 'pose.mp4']
+>>> dataset[0]["json"]
+{"label": "signing_hello"}
+```
+
 ## Lance
 
 [Lance](https://lance.org) is an open multimodal lakehouse table format. Lance tables can natively store not only text and scalar values,