Downloads · 30 days
0
shizhediao2/Llama-Nemotron-8B-v1-Prorl
Llama-Nemotron-8B-v1-Prorl is a machine learning model from shizhediao2. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Llama-Nemotron-8B-v1-Prorl is the world’s leading 8B reasoning model for complex tasks such as mathematical problems, coding challenges, and scientific questions. It builds on top of REINFORCE++-baseline with dynamic…
Downloads · 30 days
0
Access
Public
Updated Aug 25, 2025
Repo size
—
Likes
0
Public
Click a slice to open those files.
.md6.4 KB · 81%
From the Hugging Face model README
Llama-Nemotron-8B-v1-Prorl is the world’s leading 8B reasoning model for complex tasks such as mathematical problems, coding challenges, and scientific questions. It builds on top of REINFORCE++-baseline with dynamic sampling and clip-higher, and proposes several critical enhancements such as periodically refreshing the reference model with the current best checkpoint and imposing the length penalty only in scheduled cycles. Together, these techniques allow model performance to continually improve with more RL training steps and expand LLMs' reasoning boundaries. Our latest checkpoint has achieved impressive results, outperforming Deepseek’s model by a large margin on a broad range of tasks including math, coding, and GPQA.
TBD
Global <br>
Researchers and developers can use this model to solve math, coding and STEM questions.
Huggingface x/xx/2025 via [URL] <br>
Haven't published yet but will release a paper at the same time as the model release. <br>
Prorl: Prolonged reinforcement learning expands reasoning boundaries in large language models <br> Llama-Nemotron: Efficient Reasoning Models Technical Report
Architecture Type: Dense decoder-only Transformer model <br>
Network Architecture: nvidia/Llama-3.1-Nemotron-Nano-8B-v1 <br>
**This model was developed based on Llama-3.1-Nemotron-Nano-8B-v1 <br>
Input Type(s): Text <br> Input Format: String <br> Input Parameters: 1D <br> Other Properties Related to Input: Context length up to 32,000 tokens <br>
Output Type(s): Text <br> Output Format: String <br> Output Parameters: 1D <br> Other Properties Related to Output: Context length up to 32,000 tokens <br>
Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions. <br>
Runtime Engine(s): Transformers
Supported Hardware Microarchitecture Compatibility: <br>
Preferred/Supported Operating System(s):
1.0
** The total size (in number of data points): 479K <br> ** Total number of datasets: 5 <br>
** Dataset partition: Training [90%], testing [5%], validation [5%] <br> ** Time period for training data collection [1984-01-01 to 2023-01-01] <br> ** Time period for testing data collection [2024-01-01 to 2025-04-01] <br> ** Time period for validation data collection [2024-01-01 to 2025-04-01] <br>
Link:
Data Collection Method by dataset: <br>
Labeling Method by dataset: <br>
Properties (Quantity, Dataset Descriptions, Sensor(s)): 479K question and answer pairs <br>
Link:
Data Collection Method by dataset: <br>
Labeling Method by dataset: <br>
Properties (Quantity, Dataset Descriptions, Sensor(s)): 22K question and answer pairs <br>
Link: AIME: https://huggingface.co/datasets/opencompass/AIME2025
AMC: https://huggingface.co/datasets/AI-MO/aimo-validation-amc
**Benchmark Score <br>
| Dataset | Score |
|---|---|
| AIME | 65.83 |
| AMC | 89.30 |
Data Collection Method by dataset: <br>
Labeling Method by dataset: <br>
Properties (Quantity, Dataset Descriptions, Sensor(s)): 100 math question and answer pairs <br>
Acceleration Engine: Transformers <br>
Test Hardware: <br>
- 1x H100-80GB GPU
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our 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 here.