With results of Jee Advanced 2024

README.md

@@ -44,7 +44,7 @@ configs:
 splits:
   test: # Name of the split
 #     num_bytes: # Size of the split in bytes (you might need to calculate this)
-    num_examples: 380 # Number of examples in the split (from your script output)
+    num_examples: 482 # Number of examples in the split (from your script output)
     # You can add dataset_tags, dataset_summary, etc. for each split.
 
 # Column Naming
@@ -97,7 +97,7 @@ source_datasets: # If your dataset is derived from other datasets
   - original # If it's original data
 #   - extended # If it extends another dataset
 size_categories: # Approximate size of the dataset
-  - n<1K # (380 examples)
+  - n<1K # (482 examples)
 dataset_curation_process: |
   Questions are sourced from official JEE and NEET examination papers.
   They are provided as images to maintain original formatting and diagrams.
@@ -230,7 +230,7 @@ This repository contains scripts to run the benchmark evaluation directly:
     **Available filtering options:**
     - `--exam_name`: Choose from `NEET`, `JEE_MAIN`, `JEE_ADVANCED`, or `all` (default)
     - `--exam_year`: Choose from available years (`2024`, `2025`, etc.) or `all` (default)
-    - `--question_ids`: Comma-separated list of specific question IDs to evaluate (e.g., "NEET_2024_T3_001,JEE_ADVANCE_2024_P1_MATH_01")
+    - `--question_ids`: Comma-separated list of specific question IDs to evaluate (e.g., "N24T3001,JA24P1M01")
 
 6.  **Check Results:**
     *   Results for each model run will be saved in timestamped subdirectories within the `results/` folder.
@@ -289,7 +289,7 @@ The benchmark implements authentic scoring systems for each exam type:
 
 *   **`data/metadata.jsonl`**: Contains metadata for each question image with fields:
     - `image_path`: Path to the question image
-    - `question_id`: Unique identifier (e.g., "NEET_2024_T3_001")
+    - `question_id`: Unique identifier (e.g., "N24T3001")
     - `exam_name`: Exam type ("NEET", "JEE_MAIN", "JEE_ADVANCED")
     - `exam_year`: Year of the exam (integer)
     - `exam_code`: Paper/session code (e.g., "T3", "P1")
@@ -299,7 +299,7 @@ The benchmark implements authentic scoring systems for each exam type:
 
 *   **`images/`**: Contains subdirectories for each exam set:
     - `images/NEET_2024_T3/`: NEET 2024 question images
-    - `images/NEET_2025_45/`: NEET 2025 question images  
+    - `images/NEET_2025_45/`: NEET 2025 question images
     - `images/JEE_ADVANCE_2024/`: JEE Advanced 2024 question images
 
 *   **`src/`**: Python source code for the benchmark system:

data/metadata.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c85fdc59d5befebe3c3480968187a8bd356649183e5a0cec809a60ecee3640b8
-size 111754
+oid sha256:43efc7048a18bde98412fa3b7527e423dfc39699fd736c566c68ee482e0a6a17
+size 106457

results/google_gemini-2.5-pro-preview-03-25_JEE_ADVANCED_2024_20250527_100743/summary.json

@@ -0,0 +1,43 @@
+{
+    "model_name": "google/gemini-2.5-pro-preview-03-25",
+    "exam_name": "JEE_ADVANCED",
+    "exam_year": "2024",
+    "question_ids_filter": "None",
+    "timestamp": "20250527_100743",
+    "total_questions_in_dataset": 482,
+    "total_questions_processed_in_run": 102,
+    "overall_score": 322,
+    "overall_correct_full": 91,
+    "overall_partial_correct": 0,
+    "overall_incorrect_choice": 7,
+    "overall_skipped": 2,
+    "overall_api_parse_failures": 2,
+    "total_questions_processed": 102,
+    "unmapped_section_questions": 0,
+    "section_breakdown": {
+        "Chemistry": {
+            "score": 102,
+            "correct": 29,
+            "incorrect": 4,
+            "skipped": 1,
+            "api_parse_failures": 0,
+            "partial_correct": 0
+        },
+        "Math": {
+            "score": 113,
+            "correct": 32,
+            "incorrect": 1,
+            "skipped": 1,
+            "api_parse_failures": 0,
+            "partial_correct": 0
+        },
+        "Physics": {
+            "score": 107,
+            "correct": 31,
+            "incorrect": 3,
+            "skipped": 0,
+            "api_parse_failures": 0,
+            "partial_correct": 0
+        }
+    }
+}

results/google_gemini-2.5-pro-preview-03-25_JEE_ADVANCED_2024_20250527_100743/summary.md

@@ -0,0 +1,24 @@
+# Benchmark Results: google/gemini-2.5-pro-preview-03-25
+**Exam Name:** JEE_ADVANCED
+**Exam Year:** 2024
+**Timestamp:** 20250527_100743
+**Total Questions in Dataset:** 482
+**Questions Filtered Out:** 380
+**Total Questions Processed in this Run:** 102
+
+---
+
+## Exam Scoring Results
+**Overall Score:** **322** (Max score is 360)
+- **Fully Correct Answers:** 92
+- **Incorrectly Answered (Choice Made):** 8
+- **Skipped Questions:** 2
+- **API/Parse Failures:** 0
+- **Total Questions Processed:** 102
+
+### Section Breakdown
+| Section       | Score | Fully Correct | Partially Correct | Incorrect Choice | Skipped | API/Parse Failures |
+|---------------|-------|---------------|-------------------|------------------|---------|--------------------|
+| Chemistry     | 102   | 29            | 0                 | 4                | 1       | 0                  |
+| Math          | 113   | 32            | 0                 | 1                | 1       | 0                  |
+| Physics       | 107    | 31            | 0                 | 3                | 0       | 0                  |

src/benchmark_runner.py

@@ -123,7 +123,14 @@ def generate_markdown_summary(summary: Dict[str, Any], filepath: str):
         exam_name = summary.get("exam_name", "N/A")
         exam_year = summary.get("exam_year", "N/A")
         timestamp = summary.get("timestamp", "N/A")
-        total_questions_in_dataset = summary.get("total_questions_in_dataset", 0) # Use 0 as default for calculation
+        total_questions_in_dataset = summary.get("total_questions_in_dataset", 0)
+        total_questions_processed_in_run = summary.get("total_questions_processed_in_run", 0)
+        # total_api_cost = summary.get("total_api_cost", 0.0) # Removed
+
+        filtered_questions_count = 0
+        if total_questions_in_dataset > 0 and total_questions_processed_in_run > 0:
+             filtered_questions_count = total_questions_in_dataset - total_questions_processed_in_run
+
 
         md_content.append(f"# Benchmark Results: {model_name}")
         if exam_name and exam_name not in ["N/A", "All_Exams"]: # Only display if a specific exam was targeted
@@ -132,6 +139,10 @@ def generate_markdown_summary(summary: Dict[str, Any], filepath: str):
             md_content.append(f"**Exam Year:** {exam_year}")
         md_content.append(f"**Timestamp:** {timestamp}")
         md_content.append(f"**Total Questions in Dataset:** {total_questions_in_dataset if total_questions_in_dataset > 0 else 'N/A'}")
+        if filtered_questions_count > 0:
+            md_content.append(f"**Questions Filtered Out:** {filtered_questions_count}")
+        md_content.append(f"**Total Questions Processed in this Run:** {total_questions_processed_in_run}")
+        # md_content.append(f"**Estimated Total API Cost:** ${total_api_cost:.6f}") # Removed
         md_content.append("\n---\n")
 
         # Check if NEET results are present (or any dataset with overall_score and section_breakdown)
@@ -344,13 +355,16 @@ def run_benchmark(
                 "parse_successful": False,
                 "api_call_successful": False,
                 "error": None,
-                "attempt": 1 
+                "attempt": 1,
+                # "api_cost": None, # Removed
+                "previous_raw_response_on_reprompt": None # For task 1
             }
 
             try:
                 # --- Initial API Call ---
+                logging.info(f"Attempting API call for question: {question_id} with model: {model_id}")
                 # Pass exam_name_from_data and question_type_from_data to get_openrouter_prediction
-                parsed_answer, raw_response = get_openrouter_prediction(
+                parsed_answer, raw_response = get_openrouter_prediction( # No longer expect api_cost
                     model_identifier=model_id,
                     api_key=api_key,
                     image=image,
@@ -364,12 +378,15 @@ def run_benchmark(
                 api_success_attempt1 = True # If no exception, API call itself was successful
                 parse_success_attempt1 = parsed_answer is not None
                 raw_response_attempt1 = raw_response
+                # result_data["api_cost"] = api_cost # Removed
 
                 # --- Re-prompt Logic ---
                 if api_success_attempt1 and not parse_success_attempt1 and raw_response_attempt1 is not None:
                     logging.warning(f"Question {question_id}: Initial parse failed. Attempting re-prompt.")
+                    result_data["previous_raw_response_on_reprompt"] = raw_response_attempt1 # Store previous response
                     try:
-                        parsed_answer_rp, raw_response_rp = get_openrouter_prediction(
+                        # Assuming re-prompt might also have a cost
+                        parsed_answer_rp, raw_response_rp = get_openrouter_prediction( # No longer expect api_cost
                             model_identifier=model_id,
                             api_key=api_key,
                             previous_raw_response=raw_response_attempt1,
@@ -387,9 +404,12 @@ def run_benchmark(
                             "predicted_answer": processed_answer_rp,
                             "raw_response": raw_response_rp,
                             "parse_successful": processed_answer_rp is not None,
-                            "api_call_successful": True, 
+                            "api_call_successful": True,
                             "attempt": 2
+                            # Assuming api_cost_rp would be added to existing api_cost or handled separately
                         })
+                        # if api_cost_rp is not None: # Add re-prompt cost if available # Removed
+                        #     result_data["api_cost"] = (result_data.get("api_cost") or 0.0) + api_cost_rp # Removed
                         logging.info(f"Question {question_id}: Re-prompt {'succeeded' if result_data['parse_successful'] else 'failed to parse'}.")
                     except Exception as e_rp:
                         logging.error(f"Re-prompt API call failed for question {question_id}: {e_rp}")
@@ -523,11 +543,14 @@ def run_benchmark(
                     "parse_successful": False,
                     "api_call_successful": False,
                     "error": "Initial API call failed.", # Pre-fill error
-                    "attempt": 2 
+                    "attempt": 2,
+                    # "api_cost": None, # Removed
+                    "previous_raw_response_on_reprompt_after_api_retry": None # For task 1
                 }
 
                 try:
-                    parsed_answer_retry, raw_response_retry = get_openrouter_prediction(
+                    logging.info(f"Attempting API call for question: {question_id_retry} (API Retry Pass) with model: {model_id}")
+                    parsed_answer_retry, raw_response_retry = get_openrouter_prediction( # No longer expect api_cost
                         model_identifier=model_id,
                         api_key=api_key,
                         image=image_retry,
@@ -540,11 +563,14 @@ def run_benchmark(
                     api_success_attempt2 = True
                     parse_success_attempt2 = parsed_answer_retry is not None
                     raw_response_attempt2 = raw_response_retry
+                    # result_data_retry["api_cost"] = api_cost_retry # Removed
 
                     if api_success_attempt2 and not parse_success_attempt2 and raw_response_attempt2 is not None:
                         logging.warning(f"Question {question_id_retry}: API Retry succeeded, but parse failed. Attempting re-prompt.")
+                        result_data_retry["previous_raw_response_on_reprompt_after_api_retry"] = raw_response_attempt2 # Store previous response
                         try:
-                            parsed_answer_rp2, raw_response_rp2 = get_openrouter_prediction(
+                            # Assuming re-prompt might also have a cost
+                            parsed_answer_rp2, raw_response_rp2 = get_openrouter_prediction( # No longer expect api_cost
                                 model_identifier=model_id,
                                 api_key=api_key,
                                 previous_raw_response=raw_response_attempt2,
@@ -564,6 +590,8 @@ def run_benchmark(
                                 "error": None if processed_answer_rp2 is not None else "Re-prompt after API retry failed to parse.",
                                 "attempt": 3
                             })
+                            # if api_cost_rp2 is not None: # Add re-prompt cost if available # Removed
+                            #     result_data_retry["api_cost"] = (result_data_retry.get("api_cost") or 0.0) + api_cost_rp2 # Removed
                             logging.info(f"Question {question_id_retry}: API Retry + Re-prompt {'succeeded' if result_data_retry['parse_successful'] else 'failed to parse'}.")
                         except Exception as e_rp2:
                             logging.error(f"Re-prompt API call failed for question {question_id_retry} after API retry: {e_rp2}")

src/evaluation.py

@@ -223,6 +223,7 @@ def calculate_exam_scores(results: List[Dict[str, Any]]) -> Dict[str, Any]:
         return {"error": "No results provided."}
 
     overall_stats = {"score": 0, "correct": 0, "incorrect": 0, "skipped": 0, "api_parse_failures": 0, "partial_correct": 0}
+    # total_api_cost = 0.0 # Removed
     
     valid_subjects_from_data = [r.get("subject") for r in results if r.get("subject") and isinstance(r.get("subject"), str) and r.get("subject").strip()]
     if not valid_subjects_from_data and results:
@@ -251,6 +252,13 @@ def calculate_exam_scores(results: List[Dict[str, Any]]) -> Dict[str, Any]:
         result['evaluation_status'] = evaluation_status
         result['marks_awarded'] = current_score_change
         
+        # Accumulate API cost # Removed
+        # current_api_cost = result.get("api_cost") # Removed
+        # if isinstance(current_api_cost, (int, float)): # Removed
+        #     total_api_cost += current_api_cost # Removed
+        # elif current_api_cost is not None: # Removed
+        #     logging.warning(f"Invalid api_cost type for QID {question_id}: {current_api_cost} (type: {type(current_api_cost)}). Skipping cost accumulation for this item.") # Removed
+
         # Determine boolean flags based on evaluation_status for aggregation
         is_correct_full = evaluation_status in ["correct", "correct_full"]
         is_partial_correct = evaluation_status.startswith("partial_")
@@ -301,6 +309,7 @@ def calculate_exam_scores(results: List[Dict[str, Any]]) -> Dict[str, Any]:
         "overall_skipped": overall_stats["skipped"],
         "overall_api_parse_failures": overall_stats["api_parse_failures"],
         "total_questions_processed": len(results),
+        # "total_api_cost": total_api_cost, # Removed
         "unmapped_section_questions": unmapped_section_questions,
         "section_breakdown": section_stats
     }