Switch to ImageFolder format for HuggingFace dataset viewer

The JSON builder was treating image paths as local files on the
HuggingFace worker, causing FileNotFoundError. Switching to
ImageFolder format by:
- Renaming 'image' to 'file_name' in metadata.jsonl (ImageFolder convention)
- Pointing data_files to images/**/*.png instead of the JSONL
- Adding drop_labels: true to prevent auto-classification

ImageFolder properly resolves image paths via the Hub API.

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

README.md

@@ -19,7 +19,8 @@ configs:
   - config_name: default
     data_files:
       - split: test
-        path: metadata.jsonl
+        path: "images/**/*.png"
+    drop_labels: true
 dataset_info:
   features:
     - name: image
@@ -242,7 +243,7 @@ The benchmark implements authentic scoring systems for each exam type:
 ## Dataset Structure
 
 *   **`metadata.jsonl`**: Contains metadata for each question image with fields:
-    - `image`: Path to the question image
+    - `file_name`: Path to the question image (relative to repo root)
     - `question_id`: Unique identifier (e.g., "N24T3001")
     - `exam_name`: Exam type ("NEET", "JEE_MAIN", "JEE_ADVANCED")
     - `exam_year`: Year of the exam (integer)

metadata.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:54ccf60f4d0d436c23518228184ca4184942f735433383c484871ddd9a6e6324
-size 130566
+oid sha256:c08f97e33417afb8dd931439b1f4cb62aa30b8bf6e206accb5e12198b383a50c
+size 132878

src/benchmark_runner.py

@@ -483,7 +483,7 @@ def run_benchmark(
         with open(metadata_path, 'r', encoding='utf-8') as f:
             for line in f:
                 row = json.loads(line)
-                row["image"] = os.path.join(images_base_dir, row["image"])
+                row["image"] = os.path.join(images_base_dir, row.pop("file_name"))
                 row.setdefault("paper_id", None)
                 records.append(row)