Opencodereasoning Nemotron 14B

🚀 OpenCodeReasoning-Nemotron-14B

OpenCodeReasoning-Nemotron-14B is a large language model derived from Qwen2.5-14B-Instruct, post - trained for code generation reasoning, supporting a 32K token context length and suitable for both commercial and non - commercial use.

🚀 Quick Start

Run Inference on Coding Problems

import transformers
import torch

model_id = "nvidia/OpenCodeReasoning-Nemotron-14B"

pipeline = transformers.pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

prompt = """You are a helpful and harmless assistant. You should think step-by-step before responding to the instruction below.

Please use python programming language only.

You must use ```python for just the final solution code block with the following format:
```python
# Your code here

{user} """

messages = [ { "role": "user", "content": prompt.format(user="Write a program to calculate the sum of the first $N$ fibonacci numbers")}, ]

outputs = pipeline( messages, max_new_tokens=32768, ) print(outputs[0]["generated_text"][-1]['content'])

## ✨ Features
- **Based on Qwen2.5-14B-Instruct**: OpenCodeReasoning-Nemotron-14B is a derivative of Qwen2.5-14B-Instruct, inheriting its powerful language processing capabilities.
- **Code Generation Reasoning**: It is a reasoning model post - trained for code generation, enabling it to handle complex coding problems.
- **32K Token Context Length**: Supports a context length of 32K tokens, allowing for more comprehensive input and output.
- **Commercial and Non - commercial Use**: The model is ready for both commercial and non - commercial use.

## 📦 Installation
The README does not provide specific installation steps, so this section is skipped.

## 💻 Usage Examples

### Basic Usage
```python
import transformers
import torch

model_id = "nvidia/OpenCodeReasoning-Nemotron-14B"

pipeline = transformers.pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

prompt = """You are a helpful and harmless assistant. You should think step-by-step before responding to the instruction below.

Please use python programming language only.

You must use ```python for just the final solution code block with the following format:
```python
# Your code here

{user} """

messages = [ { "role": "user", "content": prompt.format(user="Write a program to calculate the sum of the first $N$ fibonacci numbers")}, ]

outputs = pipeline( messages, max_new_tokens=32768, ) print(outputs[0]["generated_text"][-1]['content'])

### Advanced Usage
There is no specific advanced usage example provided in the original README, so this part is not further expanded.

## 📚 Documentation

### Model Overview
OpenCodeReasoning-Nemotron-14B is a large language model (LLM) that is a derivative of Qwen2.5-14B-Instruct. It is a reasoning model post - trained for code generation. The model supports a context length of 32K tokens and is suitable for both commercial and non - commercial use.

### Evaluation Results
![Evaluation Results](./results.png)

The following results are the average of **64 evaluations** on each benchmark:
| Model                  | LiveCodeBench Avg. | CodeContest All |
|------------------------|--------------------|-----------------|
| DeepSeek-R1            | 65.6               | 26.2            |
| QwQ-32B                | 61.3               | 20.2            |
|                        |                    |                 |
| **Distilled 7B+ Models** |                    |                 |
|                        |                    |                 |
| Bespoke-Stratos-7B     | 14.7               | 2.0             |
| OpenThinker-7B         | 25.5               | 5.0             |
| R1-Distill-Qwen-7B     | 38.0               | 11.1            |
| OlympicCoder-7B        | 40.9               | 10.6            |
| **OCR-Qwen-7B** | **48.5** | **16.3** |
| **OCR-Qwen-7B-Instruct** | **51.3** | **18.1** |
|                        |                    |                 |
| **Distilled 14B+ Models**|                    |                 |
|                        |                    |                 |
| R1-Distill-Qwen-14B    | 51.3               | 17.6            |
| **OCR-Qwen-14B** | **57.7** | **22.6** |
| **OCR-Qwen-14B-Instruct**| **59.4** | **23.6** |
|                        |                    |                 |
| **Distilled 32B+ Models**|                    |                 |
|                        |                    |                 |
| Bespoke-Stratos-32B    | 30.1               | 6.3             |
| OpenThinker-32B        | 54.1               | 16.4            |
| R1-Distill-Qwen-32B    | 58.1               | 18.3            |
| OlympicCoder-32B       | 57.4               | 18.0            |
| **OCR-Qwen-32B** | **61.8** | **24.6** |
| **OCR-Qwen-32B-Instruct**| **61.7** | **24.4** |

### Reproducing Results
- [Models](https://huggingface.co/collections/nvidia/opencodereasoning-2-68168f37cd7c6beb1e3f92e7)
- [Dataset](https://huggingface.co/datasets/nvidia/OpenCodeReasoning)
- [Paper](https://arxiv.org/abs/2504.01943)

### Model Details
| Property | Details |
|----------|---------|
| Model Type | Dense decoder - only Transformer model, based on Qwen-14B-Instruct |
| Training Data | [OpenCodeReasoning](https://huggingface.co/datasets/nvidia/OpenCodeReasoning) dataset, composed of competitive programming questions and DeepSeek - R1 generated responses, with 736k samples |
| Input Type | Text |
| Input Format | String |
| Input Parameters | One - Dimensional (1D), context length up to 32,768 tokens |
| Output Type | Text |
| Output Format | String |
| Output Parameters | One - Dimensional (1D), context length up to 32,768 tokens |
| Runtime Engine | NeMo 2.3.0 |
| Recommended Hardware | NVIDIA Ampere, NVIDIA Hopper |
| Preferred OS | Linux |
| Model Version | 1.0 (4/25/2025), including OpenCodeReasoning - Nemotron - 7B, OpenCodeReasoning - Nemotron - 14B, OpenCodeReasoning - Nemotron - 32B, OpenCodeReasoning - Nemotron - 32B - IOI |
| Inference Engine | vLLM |
| Test Hardware | NVIDIA H100 - 80GB |

### License
Use of this model is governed by [Apache 2.0](https://huggingface.co/nvidia/OpenCode-Nemotron-2-14B/blob/main/LICENSE).

### Deployment
- **Deployment Geography**: Global
- **Use Case**: Intended for developers and researchers building LLMs.
- **Release Date**: Huggingface [04/25/2025] via https://huggingface.co/nvidia/OpenCodeReasoning-Nemotron-7B/

### Reference
[2504.01943] OpenCodeReasoning: Advancing Data Distillation for Competitive Coding

### Ethical Considerations
NVIDIA believes Trustworthy AI is a shared responsibility and has established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with the terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse. Please report security vulnerabilities or NVIDIA AI Concerns.

## 🔧 Technical Details
The model is developed based on Qwen2.5-14B-Instruct and has 14B model parameters. It uses a dense decoder - only Transformer architecture. The training data is the [OpenCodeReasoning](https://huggingface.co/datasets/nvidia/OpenCodeReasoning) dataset, which combines competitive programming questions and DeepSeek - R1 generated responses.

## 📄 License
The model is licensed under the [Apache 2.0](https://huggingface.co/nvidia/OpenCode-Nemotron-2-14B/blob/main/LICENSE) license.

## Citation
If you find the data useful, please cite:

@article{ahmad2025opencodereasoning, title={OpenCodeReasoning: Advancing Data Distillation for Competitive Coding}, author={Wasi Uddin Ahmad, Sean Narenthiran, Somshubra Majumdar, Aleksander Ficek, Siddhartha Jain, Jocelyn Huang, Vahid Noroozi, Boris Ginsburg}, year={2025}, eprint={2504.01943}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2504.01943}, }