Add dataset card for LongVLBench
#1
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
task_categories:
|
| 3 |
+
- image-text-to-text
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# LongVLBench
|
| 7 |
+
|
| 8 |
+
LongVLBench is a long-context multimodal reasoning benchmark introduced in the paper [VLZip: Unified Visual and Textual Compression for Interleaved Long-Context Modeling](https://huggingface.co/papers/2608.08630). It is derived from video narratives and demands holistic, narrative-level reasoning over long, interleaved image-text sequences.
|
| 9 |
+
|
| 10 |
+
This dataset is used to evaluate vision-language models on long-context multimodal tasks. For more details, please refer to the [paper](https://huggingface.co/papers/2608.08630) and the [GitHub repository](https://github.com/ShareLab-SII/VLZip).
|