Update data files and README with latest annotations

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

README.md

@@ -57,7 +57,7 @@ configs:
 
 | Metric | Full Version | Lite Version |
 |--------|--------------|--------------|
-| 📊 **Total Questions** | 27,936 five-way multiple-choice | 5,960 questions (21.3%) |
+| 📊 **Total Questions** | 27,631 five-way multiple-choice | 2,000 questions (7.2%) |
 | 🎬 **Videos** | 335 unique videos | 68 unique videos (20.3%) |
 | ⏱️ **Avg. Video Length** | 424 seconds | 421 seconds |
 | 📁 **Total Size** | ~161 GB | ~35 GB |
@@ -76,12 +76,12 @@ Current video QA benchmarks often allow models to answer questions using single
 HERBench is available in two versions to accommodate different storage and computational constraints:
 
 #### Full Version (~161 GB)
-- **27,936 questions** across **335 videos**
+- **27,631 questions** across **335 videos**
 - Complete benchmark for comprehensive evaluation
 - Recommended for: Final paper results, thorough model evaluation, benchmarking
 
 #### Lite Version (~35 GB) 🚀
-- **5,960 questions** across **68 videos** (21.3% subset)
+- **2,000 questions** across **68 videos** (7.2% subset)
 - Same task distribution and difficulty as full version
 - Videos sampled to maintain diversity across all 12 tasks
 - Recommended for: Quick prototyping, limited storage, initial experiments, development
@@ -200,8 +200,8 @@ Videos are sourced from diverse, high-quality datasets:
 ```
 HERBench/
 ├── data/
-│   ├── herbench_annotations.json        # Full: 27,936 questions
-│   ├── herbench_annotations_lite.json   # Lite: ~5,600 questions
+│   ├── herbench_annotations.json        # Full: 27,631 questions
+│   ├── herbench_annotations_lite.json   # Lite: 2,000 questions
 │   ├── task_metadata.json               # Task descriptions (shared)
 │   ├── video_metadata.json              # Video information (shared)
 │   └── README_DATA.md                   # Data format documentation
@@ -268,10 +268,10 @@ For detailed format documentation, see [data/README_DATA.md](data/README_DATA.md
 # Install Hugging Face CLI
 pip install huggingface-hub
 
-# Download FULL version (27,936 questions, ~161 GB)
+# Download FULL version (27,631 questions, ~161 GB)
 huggingface-cli download DanBenAmi/HERBench --repo-type dataset --local-dir HERBench
 
-# Download LITE version only (~5,600 questions, ~35 GB videos)
+# Download LITE version only (2,000 questions, ~35 GB videos)
 huggingface-cli download DanBenAmi/HERBench \
     --include "data/herbench_lite.parquet" \
     --include "data/*metadata.json" \
@@ -294,7 +294,7 @@ The dataset is provided in **Parquet format** for optimal compatibility with Hug
 ```python
 from datasets import load_dataset
 
-# Load FULL version (default) - 27,936 questions
+# Load FULL version (default) - 27,631 questions
 dataset_full = load_dataset("DanBenAmi/HERBench", "full")
 print(f"Total questions: {len(dataset_full['test'])}")
 
@@ -309,7 +309,7 @@ print(f"Answer: {example['answer']}")
 print(f"Task: {example['task_type']}")
 print(f"Video: {example['video_path']}")
 
-# Load LITE version - ~5,600 questions (20% sample)
+# Load LITE version - 2,000 questions (7.2% sample)
 dataset_lite = load_dataset("DanBenAmi/HERBench", "lite")
 print(f"Lite questions: {len(dataset_lite['test'])}")
 ```
@@ -368,6 +368,40 @@ rm videos_lite.tar
 
 > **Note:** The archive is structured so lite videos are in the first parts (00-03). This means if you download the full version, you automatically have the lite videos too!
 
+#### Troubleshooting Extraction Issues
+
+If you encounter `"This does not look like a tar archive"` or other errors:
+
+1. **Verify checksums** to check download integrity:
+   ```bash
+   cd HERBench/videos
+   sha256sum -c videos.tar.checksums.txt
+   ```
+
+2. **Check file format** of downloaded parts:
+   ```bash
+   file videos.tar.part.00
+   # Expected: "POSIX tar archive (GNU)" or similar
+   ```
+
+3. **Test before full extraction**:
+   ```bash
+   cat videos.tar.part.{00..03} > videos_lite.tar
+   tar -tvf videos_lite.tar | head -10
+   # Should list video files like: HD_EPIC/P01/P01-*.mp4
+   ```
+
+4. **Re-download** specific parts if checksums fail:
+   ```bash
+   huggingface-cli download DanBenAmi/HERBench \
+       --repo-type dataset \
+       --include "videos/videos.tar.part.00" \
+       --local-dir HERBench \
+       --force-download
+   ```
+
+If issues persist, please [open an issue on GitHub](https://github.com/DanBenAmi/HERBench/issues).
+
 ### 3. Load and Use the Data
 
 ```python
@@ -477,7 +511,8 @@ This work was supported by [Institution/Grant acknowledgments to be added].
 
 ## 🔄 Updates
 
-- **v1.0.0** (January 2025): Initial release with 27,936 questions across 335 videos
+- **v1.0.0** (February 2026): Cleaned dataset — removed 305 questions with corrupted VLM descriptions; Full: 27,631 questions, Lite: 2,000 questions (stratified subsample)
+
 
 
 ---

data/herbench_annotations.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02f66afb2780851304069526e480414d74cc78f30554ac38465a89880f9ad799
-size 37911086
+oid sha256:91c5eec5bd74d8966259addb736eb76e997148138e5a567bb8f5d02219d419dd
+size 37470860

data/herbench_annotations_lite.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:24c5d6806ba159d9e57543e631f66153ea41f07172ee316fc71d41bb5b31e1f0
-size 8113978
+oid sha256:dce4898dcfd87be7a02a8f5f3baf0add4a4a8ba3413d09a2a7634c94613597e6
+size 2719239

data/herbench_full.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65521ae94315ec5804b74addeb694a3bccbc9b3796fdc24fcb657212e9b2c561
-size 5009650
+oid sha256:94ee0876345c3a173941bd48ee3b98169eac96fb0ccf2d0d1017c9914992de56
+size 4984150

data/herbench_lite.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63708b468e77c1e582fdcb056d21ad357ab9d89b3b2b84cc44d476b62ce96b33
-size 1072889
+oid sha256:d0bbefbda70258cb5d826d23d96704081af95577e4cb49ae23f76367f8eaf13d
+size 455813

data/task_metadata.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c5904d02b589338326345bed64e739dcab9b5155a2875b09c583a209aaacefa
+oid sha256:d6fcb60810e9711b15ef5de88cf71a7556de3a6883e9a78c5d7e2ff6a8972d7d
 size 12614

data/video_metadata.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:89929264e403bc57876cf8cda83321ee61294064e2a348c48e44185bfa54fde6
-size 102696
+oid sha256:4572a37586c35709d8b79de0c3254bc4a768e79f567a98df3983ec904a415ead
+size 102695