---
pretty_name: AMC 2023 (RLVR format)
license: other
license_name: maa-competition-problems
license_link: https://maa.org/
language:
- en
task_categories:
- text-generation
- question-answering
tags:
- math
- reasoning
- rlvr
- competition-math
- amc
- evaluation
size_categories:
- n<1K
configs:
- config_name: default
  data_files:
  - split: test
    path: test.parquet
---

# AMC 2023 (RLVR format)

This dataset contains **40 problems from the 2023 AMC 12 (AMC 12A and AMC 12B)**, each with its final answer, in the verl-style RLVR (reinforcement learning with verifiable rewards) format. In our paper *Adaptive Negative Reinforcement for LLM Reasoning: Dynamically Balancing Correction and Diversity in RLVR*, it is one of our three evaluation benchmarks, of moderate difficulty between MATH and AIME 2025.

The file is an unmodified copy of the AMC23 evaluation data released by Zhu et al. with *The Surprising Effectiveness of Negative Reinforcement in LLM Reasoning* (NeurIPS 2025). The `data_source` field is `TianHongZXY/amc23`. We re-host the data so that our experiments can be reproduced with exactly the same file.

> **Evaluation only.** This dataset is intended for evaluation. Do not train on it.

## Dataset summary

| Split | Problems | Use in our paper |
|-------|---------:|------------------|
| `test` | 40 | Evaluation (Pass@k, k = 1 … 256, 256 samples per problem) |

| Source contest | Problems |
|----------------|---------:|
| 2023 AMC 12A | 22 |
| 2023 AMC 12B | 18 |

The original AMC contests are multiple choice. Here each problem is posed **without the answer choices**, and the model must produce the numerical answer itself.

Because there are only 40 problems, one problem corresponds to 2.5 percentage points of Pass@k.

## Data fields

| Field | Type | Description |
|-------|------|-------------|
| `id` | int | Problem id (0–39) |
| `url` | string | Link to the problem on the Art of Problem Solving wiki |
| `data_source` | string | Always `"TianHongZXY/amc23"` |
| `prompt` | list of `{role, content}` | Chat-format prompt: one `user` message containing the problem statement |
| `ability` | string | Always `"math"` |
| `reward_model.style` | string | Always `"rule"` (rule-based verifiable reward) |
| `reward_model.ground_truth.question` | string | Problem statement |
| `reward_model.ground_truth.target` | string | Final answer |
| `extra_info.index` | int | Index within the split |
| `extra_info.split` | string | Always `"test"` |

No reference solutions are included. Solutions can be found through the `url` field.

### Example

```json
{
  "id": 0,
  "url": "https://artofproblemsolving.com/wiki/index.php/2023_AMC_12A_Problems/Problem_1",
  "data_source": "TianHongZXY/amc23",
  "prompt": [{"role": "user", "content": "Cities $A$ and $B$ are $45$ miles apart. Alicia lives in $A$ and Beth lives in $B$. Alicia bikes towards $B$ at 18 miles per hour. Leaving at the same time, Beth bikes toward $A$ at 12 miles per hour. How many miles from City $A$ will they be when they meet?"}],
  "ability": "math",
  "reward_model": {
    "style": "rule",
    "ground_truth": {
      "question": "Cities $A$ and $B$ are $45$ miles apart. ...",
      "target": "27"
    }
  },
  "extra_info": {"index": 0, "split": "test"}
}
```

## Usage

```python
from datasets import load_dataset

ds = load_dataset("<your-username>/AMC23-RLVR", split="test")
for row in ds:
    question = row["prompt"][0]["content"]
    answer = row["reward_model"]["ground_truth"]["target"]
```

In our experiments with `Qwen2.5-Math-1.5B`, the problem was wrapped in this chat template:

```
<|im_start|>system
You are a helpful assistant.<|im_end|>
<|im_start|>user
{input}
Please reason step by step, and put your final answer
within \boxed{}.<|im_end|>
<|im_start|>assistant
```

Generation settings: temperature 0.6, top-p 0.95, 256 samples per problem. Pass@k was computed with the unbiased estimator of Chen et al. (2021).

## Source and license

- **Problems:** 2023 AMC 12A and AMC 12B, © Mathematical Association of America (MAA). The problems are reproduced here for non-commercial research and evaluation. Copyright remains with the MAA.
- **Problem text and links:** Art of Problem Solving wiki.
- **Preprocessing:** Zhu et al. (2025), <https://github.com/TianHongZXY/RLVR-Decomposed>.

## Citation

```bibtex
@inproceedings{zhu2025nsr,
  title     = {The Surprising Effectiveness of Negative Reinforcement in {LLM} Reasoning},
  author    = {Zhu, Xinyu and Xia, Mengzhou and Wei, Zhepei and Chen, Wei-Lin and Chen, Danqi and Meng, Yu},
  booktitle = {NeurIPS},
  year      = {2025}
}
```

Our paper:

<!-- Replace with the final citation once the paper is published. -->
```bibtex
@article{ingle2026adaptive,
  title  = {Adaptive Negative Reinforcement for {LLM} Reasoning: Dynamically Balancing Correction and Diversity in {RLVR}},
  author = {Ingle, Yash and Chauhan, Jaival and Yadav, Ankit and Mishra, Sudhakar},
  year   = {2026}
}
```

Code: [A-NSR](https://github.com/Jaival111/Adaptive-Negative-Reinforcement) · [CW-NSR](https://github.com/Jaival111/Confidence-Weighted-NSR)