---
license: mit
tags:
- degen_test_1_countdown
- stage_02_alternative
- llm_extraction
- math_spans
---

# test-updated-extractor-v3

LLM-based math span extraction with canonicalization

## Dataset Info

- **Rows**: 1
- **Columns**: 26

## Columns

| Column | Type | Description |
|--------|------|-------------|
| question | Value('string') | *No description provided* |
| metadata | Value('string') | *No description provided* |
| task_source | Value('string') | *No description provided* |
| formatted_prompt | List({'content': Value('string'), 'role': Value('string')}) | *No description provided* |
| responses_by_sample | List(List(Value('string'))) | *No description provided* |
| tokens_by_sample | List(Value('int64')) | *No description provided* |
| num_rounds_by_sample | List(Value('int64')) | *No description provided* |
| continuation_prompts_by_sample | List(List(Value('string'))) | *No description provided* |
| continuation_text_by_sample | List(List(Value('string'))) | *No description provided* |
| strategy_type | Value('string') | *No description provided* |
| termination_reason_by_sample | List(Value('string')) | *No description provided* |
| correct_round_by_sample | List(Value('null')) | *No description provided* |
| smart_termination_enabled | Value('bool') | *No description provided* |
| retry_count_by_sample | List(Value('int64')) | *No description provided* |
| answers_by_sample | List(List(List(Value('string')))) | *No description provided* |
| evals_by_sample | List(List(List(Value('bool')))) | *No description provided* |
| eval_metadata_by_sample | List(List(List({'answer_block': Value('string'), 'error': Value('null'), 'final_answer': Value('float64'), 'is_correct': Value('bool')}))) | *No description provided* |
| annotations_by_sample | List(List(Value('null'))) | *No description provided* |
| arithmetic_attempts_by_sample | List(List({'canonical_form': Value('string'), 'canonical_key': Value('string'), 'end_char': Value('int64'), 'end_char_in_round': Value('null'), 'has_free_variables': Value('bool'), 'is_legal': Value('null'), 'is_multi_operand': Value('bool'), 'is_valid_arithmetic': Value('bool'), 'match_type': Value('null'), 'offset_found': Value('bool'), 'operand_a': Value('float64'), 'operand_a_str': Value('null'), 'operand_b': Value('float64'), 'operand_b_str': Value('null'), 'operator': Value('string'), 'operator_raw': Value('null'), 'result': Value('float64'), 'result_str': Value('null'), 'round_idx': Value('int64'), 'separator_raw': Value('null'), 'span_text': Value('string'), 'start_char': Value('int64'), 'start_char_in_round': Value('null')})) | Extracted math operations with spans (compatible with stage_02a) |
| canonicalized_equations_by_sample | List(List(Value('string'))) | LLM-generated canonical equation forms |
| extraction_stats_by_sample | List({'has_free_variables': Value('int64'), 'invalid_arithmetic': Value('int64'), 'merged_text_length': Value('int64'), 'multi_operand': Value('int64'), 'offsets_found': Value('int64'), 'total_attempts_created': Value('int64'), 'total_spans_from_llm': Value('int64'), 'unevaluable': Value('int64'), 'unique_canonical_keys': Value('int64'), 'valid_arithmetic': Value('int64')}) | Statistics about LLM extraction per sample |
| expression_attempts_by_sample | List(List(Value('null'))) | *No description provided* |
| subgoals_by_sample | List(List(Value('null'))) | *No description provided* |
| failure_markers_by_sample | List(List(Value('null'))) | *No description provided* |
| subgoal_windows_by_sample | List(List(Value('null'))) | *No description provided* |
| branches_by_sample | List(List(Value('null'))) | *No description provided* |


## Generation Parameters

```json
{
  "script_name": "stage_02_alternative_llm_extractor.py",
  "model": "gpt-4o-mini",
  "hyperparameters": {
    "chunking_mode": "paragraph"
  },
  "input_datasets": [
    "reasoning-degeneration-dev/EAS-Qwen3-30B-A3B-Thinking-2507-cd8arg-iterative"
  ],
  "description": "LLM-based math span extraction with canonicalization"
}
```

## Experiment Documentation

For complete experiment details, see [https://github.com/reasoning-degeneration/ReasoningDegeneration/tree/main/experiments/degen_test_1_countdown/documentation](https://github.com/reasoning-degeneration/ReasoningDegeneration/tree/main/experiments/degen_test_1_countdown/documentation)

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("reasoning-degeneration-dev/test-updated-extractor-v3", split="train")
print(f"Loaded {len(dataset)} rows")
```

---

*This dataset is tracked in [reasoning-degeneration-dev/PROJECT-MANIFEST](https://huggingface.co/datasets/reasoning-degeneration-dev/PROJECT-MANIFEST)*