Update README with correct structure and size info

README.md

@@ -60,14 +60,14 @@ Each video's specific license is recorded in the `licenseurl` and `license_type`
 ## Dataset Structure
 
 The dataset uses VideoFolder format:
-- `metadata.jsonl` - Contains all video metadata
-- `data/` - Contains all video files
+- `data/metadata.jsonl` - Video metadata
+- `data/*.mp4` - 1,902 video files (~143GB total)
 
 ### Metadata Fields
 
 | Field | Description |
 |-------|-------------|
-| `file_name` | Path to video file (e.g., `data/video.mp4`) |
+| `file_name` | Video filename (e.g., `video.mp4`) |
 | `identifier` | Internet Archive identifier |
 | `title` | Video title |
 | `description` | Full description |
@@ -93,13 +93,16 @@ The dataset uses VideoFolder format:
 
 ### Loading with Hugging Face Datasets
 
+> **Note:** Requires `torchcodec` to decode videos: `pip install torchcodec`
+
 ```python
 from datasets import load_dataset
 
-# Streaming mode (recommended for large datasets)
+# Streaming mode (recommended due to 143GB size)
 ds = load_dataset("davanstrien/prelinger-archives-open", split="train", streaming=True)
 for sample in ds:
     print(sample["title"])
+    print(sample["description"])
     break
 ```