Llama-3.3-Nemotron-70B-Reward-Multilingual

Model Overview

Description:

Llama-3.3-Nemotron-70B-Reward-Multilingual is a large language model that leverages Meta-Llama-3.3-70B-Instruct as the foundation and is fine-tuned using scaled Bradley-Terry modeling to predict the quality of LLM generated responses.

Given a multilingual conversation with multiple turns between user and assistant (of up to 4,096 tokens), it rates the quality of the final assistant turn using a reward score.

For the same prompt, a response with higher reward score has higher quality than another response with a lower reward score, but the same cannot be said when comparing the scores between responses to different prompts.

As of 15 May 2025, this model achieves the highest on RM-Bench at 82.4% and second highest JudgeBench at 69.4% among Bradley-Terry Reward Models.

See details on how this model was trained at https://arxiv.org/abs/2505.11475

This model is ready for commercial/non-commercial use.

License/Terms of Use:

GOVERNING TERMS: Use of this model is governed by the NVIDIA Open Model License . Additional Information: Llama 3.3 Community License Agreement. Built with Llama.

Deployment Geography

Global

Use Case:

Llama-3.3-Nemotron-70B-Reward-Multilingual labels an LLM-generated response to a user query with a reward score.

Release Date:

HuggingFace 06/27/2025 via https://huggingface.co/nvidia/Llama-3.3-Nemotron-70B-Reward-Multilingual

References:

• HelpSteer3-Preference
• HelpSteer2-Preference
• SteerLM method
• HelpSteer
• HelpSteer2
• The future of AI: Built with Llama
• Meta's Llama 3.3 Webpage
• Meta's Llama 3.3 Model Card

RM-Bench LeaderBoard

As of 15 May 2025, our reward models trained with HelpSteer3-Preference are the top performing Bradley-Terry reward models on RM-Bench, an improved variant of RewardBench for evaluating Reward Models in Chat, Math, Code and Safety.

Note that Skywork-Reward-Llama-3.1-8B was the best performing reward model reported on RM-Bench and we evaluated all other models.

JudgeBench LeaderBoard

As of 15 May 2025, our reward models trained with HelpSteer3-Preference are the top performing Bradley-Terry reward models on JudgeBench, a popular benchmark for evaluating LLM-as-a-judge applications relating to General Knowledge, Logical Reasoning, Math and Coding.

Note that Skywork-Reward-Gemma-2-27B was the best performing reward model reported on JudgeBench and we evaluated all other models.

Model Architecture:

Architecture Type: Transformer <br> Network Architecture: Llama 3.3 <br>

We developed this model using Llama-3.3-70B-Instruct as its foundation. This model contains 70 billion parameters.

Input:

Input Type(s): Text <br> Input Format: String <br> Input Parameters: One Dimensional (1D) <br> Other Properties Related to Input: Max of 128k tokens (but trained only on conversations up to 4K tokens)<br>

Output:

Output Type(s): Float <br> Output Format: One Single Float <br> Output Parameters: One-Dimensional (1D) <br> Other Properties Related to Output: The float value represents the quality of the response, with a higher value representing higher quality. <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>

Software Integration:

Runtime Engine(s): <br>

• [NeMo - 24.05.llama.3.1] <br>

Supported Hardware Microarchitecture Compatibility: <br>

• NVIDIA Ampere <br>
• NVIDIA Hopper <br>
• NVIDIA Turing <br>

Supported Operating System(s): Linux <br>

Quick Start

You can use the model using HuggingFace Transformers library with 2 or more 80GB GPUs (NVIDIA Ampere or newer) with at least 150GB of free disk space to accomodate the download.

This code has been tested on Transformers v4.45.0, torch v2.3.0a0+40ec155e58.nv24.3 and 2 H100 80GB GPUs, but any setup that supports meta-llama/Llama-3.1-70B-Instruct should support this model as well. If you run into problems, you can consider doing pip install -U transformers.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "nvidia/Llama-3.3-Nemotron-70B-Reward-Multilingual"

model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "What is 1+1?"
good_response = "1+1=2"
bad_response = "1+1=3"

for response in [good_response, bad_response]:
    messages = [{'role': "user", "content": prompt}, {'role': "assistant", "content": response}]
    tokenized_message = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=False, return_tensors="pt", return_dict=True)
    response_token_ids = model.generate(tokenized_message['input_ids'].cuda(),attention_mask=tokenized_message['attention_mask'].cuda(),  max_new_tokens=1, return_dict_in_generate=True, output_scores=True)
    reward = response_token_ids['scores'][0][0][0].item()
    print(reward)

# Example quality - note that higher scores means higher quality, and scores can be negative.

# reward for good_response = 6.46875
# reward for bad_response = -1.8828125

Model Version:

v1.0

Training, Testing and Evaluation Datasets:

Training Datasets:

Dataset Name: HelpSteer3 <br> Dataset Link: https://huggingface.co/datasets/nvidia/HelpSteer3

Data Collection Method by dataset <br>

• [Hybrid: Human, Synthetic] <br>

Labeling Method by dataset <br>

• [Human] <br>

Properties: <br>

• 7,660 prompts, each with a pair of responses as well as human preferences between the pair of responses.

Testing Datasets:

Dataset Name: HelpSteer3 <br> Dataset Link: https://huggingface.co/datasets/nvidia/HelpSteer3

Data Collection Method by dataset <br>

• [Hybrid: Human, Synthetic] <br>

Labeling Method by dataset <br>

• [Human] <br>

Properties: <br>

• 403 prompts, each with a pair of responses as well as human preferences between the pair of responses.

Evaluation Datasets

Dataset Name: RM-Bench <br> Dataset Link: https://huggingface.co/datasets/THU-KEG/RM-Bench

Data Collection Method by dataset <br>

• [Hybrid: Human, Synthetic] <br>

Labeling Method by dataset <br>

• [Hybrid: Human, Synthetic] <br>

Properties: <br>

• 1,327 prompts, each with three pairs of responses as well as preferences between the pair of responses.

Dataset Name: JudgeBench <br> Dataset Link: https://huggingface.co/datasets/ScalerLab/JudgeBench

Data Collection Method by dataset <br>

• [Hybrid: Human, Synthetic] <br>

Labeling Method by dataset <br>

• [Hybrid: Human, Synthetic] <br>

Properties: <br>

• 350 prompts, each with a pair of responses as well as preferences between the pair of responses.

Inference:

Engine: PyTorch <br> Test Hardware: H100, A100 80GB, A100 40GB <br>

Ethical Considerations:

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 supporting model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse. For more detailed information on ethical considerations for this model, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards.

Please report security vulnerabilities or NVIDIA AI Concerns here.

Citation

If you find this model useful, please cite the following work:

@misc{wang2025helpsteer3preferenceopenhumanannotatedpreference,
      title={Help{S}teer3-{P}reference: Open Human-Annotated Preference Data across Diverse Tasks and Languages},
      author={Zhilin Wang and Jiaqi Zeng and Olivier Delalleau and Hoo-Chang Shin and Felipe Soares and Alexander Bukharin and Ellie Evans and Yi Dong and Oleksii Kuchaiev},
      year={2025},
      eprint={2505.11475},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.11475}, 
}