Move metadata.jsonl to repo root for HuggingFace image resolution

HuggingFace resolves image paths relative to the metadata file's
directory. With metadata.jsonl in data/, image paths like
"images/NEET_2024_T3/..." resolved to "data/images/..." which
doesn't exist, causing FileNotFoundError in the dataset viewer.

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

.gitattributes

@@ -1,4 +1,4 @@
 images/** filter=lfs diff=lfs merge=lfs -text
-data/metadata.jsonl filter=lfs diff=lfs merge=lfs -text
+metadata.jsonl filter=lfs diff=lfs merge=lfs -text
 images/**/*.png filter=lfs diff=lfs merge=lfs -text
 *.tar.gz filter=lfs diff=lfs merge=lfs -text

CLAUDE.md

@@ -40,8 +40,8 @@ python src/llm_interface.py  # API calls (requires .env and network)
 ```
 benchmark_runner.py  ─── orchestrator / entry point
   ├── loads config from configs/benchmark_config.yaml
-  ├── loads dataset directly from data/metadata.jsonl (JSONL → HuggingFace Dataset)
-  │     ├── data/metadata.jsonl (question metadata, 578 questions)
+  ├── loads dataset directly from metadata.jsonl (JSONL → HuggingFace Dataset)
+  │     ├── metadata.jsonl (question metadata, 578 questions)
   │     └── images/ (question PNGs, stored in Git LFS)
   ├── calls llm_interface.py for each question
   │     ├── prompts.py (prompt templates)
@@ -55,7 +55,7 @@ benchmark_runner.py  ─── orchestrator / entry point
 
 ### Key data flow
 
-1. Dataset loaded directly from `data/metadata.jsonl` into a HuggingFace `Dataset` object, filtered by exam/year
+1. Dataset loaded directly from `metadata.jsonl` into a HuggingFace `Dataset` object, filtered by exam/year
 2. Each question image is base64-encoded and sent to OpenRouter with a structured prompt
 3. If the response can't be parsed, a re-prompt is sent (text-only, with the bad response)
 4. If the API call fails, the question is queued for retry (up to 3 attempts, exponential backoff via `tenacity`)
@@ -71,7 +71,7 @@ benchmark_runner.py  ─── orchestrator / entry point
 
 ## Important Notes
 
-- **Git LFS**: Images and `data/metadata.jsonl` are in LFS. Run `git lfs pull` after cloning.
+- **Git LFS**: Images and `metadata.jsonl` are in LFS. Run `git lfs pull` after cloning.
 - **Working directory**: Scripts must be run from project root — config, data, and image paths are resolved relative to cwd.
 - **Python 3.10+**: Uses union type syntax (`list[str] | str | None`).
 - **Models**: Configured in `configs/benchmark_config.yaml` under `openrouter_models`. All must support vision input.

README.md

@@ -19,7 +19,7 @@ configs:
   - config_name: default
     data_files:
       - split: test
-        path: data/metadata.jsonl
+        path: metadata.jsonl
 dataset_info:
   features:
     - name: image
@@ -241,7 +241,7 @@ The benchmark implements authentic scoring systems for each exam type:
 
 ## Dataset Structure
 
-*   **`data/metadata.jsonl`**: Contains metadata for each question image with fields:
+*   **`metadata.jsonl`**: Contains metadata for each question image with fields:
     - `image`: Path to the question image
     - `question_id`: Unique identifier (e.g., "N24T3001")
     - `exam_name`: Exam type ("NEET", "JEE_MAIN", "JEE_ADVANCED")

configs/benchmark_config.yaml

@@ -15,7 +15,7 @@ openrouter_models:
   # - "anthropic/claude-3-haiku"
 
 # Path to the metadata JSONL file and base directory for image paths.
-metadata_path: "data/metadata.jsonl"
+metadata_path: "metadata.jsonl"
 images_base_dir: "."
 
 # Base directory where results for each model will be saved.

src/benchmark_runner.py

@@ -476,7 +476,7 @@ def run_benchmark(
     os.makedirs(base_output_dir, exist_ok=True)
 
      # Load dataset directly from metadata.jsonl and images/
-    metadata_path = config.get("metadata_path", "data/metadata.jsonl")
+    metadata_path = config.get("metadata_path", "metadata.jsonl")
     images_base_dir = config.get("images_base_dir", ".")
     try:
         records = []
@@ -501,7 +501,7 @@ def run_benchmark(
         logging.info(f"Dataset loaded successfully from {metadata_path}. Original number of questions: {len(dataset)}")
     except Exception as e:
         logging.error(f"Failed to load dataset from '{metadata_path}': {e}")
-        logging.error("Ensure 'data/metadata.jsonl' exists and image paths are valid.")
+        logging.error("Ensure 'metadata.jsonl' exists and image paths are valid.")
         return
 
     # Filter dataset based on choices
@@ -736,7 +736,7 @@ if __name__ == "__main__":
     # Assuming benchmark_config.yaml is in a 'configs' directory relative to script or a fixed path
     # Assuming metadata.jsonl is in a 'data' directory relative to script or a fixed path
     default_config_path = "configs/benchmark_config.yaml"
-    default_metadata_path = "data/metadata.jsonl"
+    default_metadata_path = "metadata.jsonl"
 
     available_models = get_available_models(default_config_path)
     available_exam_names, available_exam_years = get_available_exam_details(default_metadata_path)