Model Overview

Description:

The NVIDIA Llama 4 Maverick 17B 128E Eagle3 BF16 model is a natively multimodal model that enables text and multimodal experiences. The model is the Eagle head of Meta's Llama 4 Maverick 17B 128E model, which is an auto-regressive language model that uses a mixture-of-experts (MoE) architecture and incorporates early fusion for native multimodality. For more information, please check here. The NVIDIA Llama 4 Maverick 17B 128E Eagle3 BF16 model incorporates Eagle speculative decoding with TensorRT Model Optimizer.

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

Third-Party Community Consideration]

This base model has been developed and built to a third-party’s requirements for this application and use case; see link to Non-NVIDIA (Llama-4-Maverick-17B-128E) Model Card.

License/Terms of Use:

llama4

Deployment Geography:

Global

Use Case:

Developers designing AI Agent systems, chatbots, RAG systems, and other AI-powered applications. Also suitable for typical instruction-following tasks.

Release Date:

Build.Nvidia.com: May 15th, 2025 via [URL]
Huggingface: May 15th, 2025 via [https://huggingface.co/nvidia/Llama-4-Maverick-17B-128E-Eagle3]

Model Architecture:

Architecture Type: Transformers
Network Architecture: Llama4

Input:

Input Type(s): Multilingual text and up to 5 images
Input Format(s): String, Images
Input Parameters: 1D, 2D
Other Properties Related to Input: Context length up to 1M

Output:

Output Type(s): Multilingual text and code
Output Format: String
Output Parameters: 1D
Other Properties Related to Output: Context length up to 1M

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.

Software Integration :

Supported Runtime Engine(s):

Tensor(RT)-LLM
SGLang

Supported Hardware Microarchitecture Compatibility:

NVIDIA Blackwell
NVIDIA Hopper
NVIDIA Lovelace

[Preferred/Supported] Operating System(s):

Linux

Model Version(s):

v0.29.0

Training and Evaluation Datasets:

** The total size (in number of data points) 503.3K
** Total number of datasets 2

** Dataset partition: Training 100%

Training Dataset:

Link: ultrachat_200k and Magpie-Llama-3.1-Pro-300K-Filtered, used for data synthesis, which is then used to train the Eagle modules. Click the links above for more information regarding the dataset.
** Data Collection Method by dataset

Hybrid: Synthetic/Human/Automated
** Labeling Method by dataset
Hybrid: Synthetic, Human, Automated
Properties: 500K samples, majority synthetic, others sourced from commercially-friendly datasets.

Evaluation Dataset:

Link: MTBench, for more details, see here
** Data Collection Method by dataset

Hybrid: Human, Synthetic
** Labeling Method by dataset
Hybrid: Human, Synthetic
Properties: 3,300 multi-turn dialogue sequences, each annotated with expert preference votes.

Inference:

Engine: Tensor(RT)-LLM or SGLang
Test Hardware: H100

Eagle Speculative Decoding

Synthesized data was obtained from Meta's Llama 4 Maverick 17B 128E model, which is then used to finetune the Eagle modules. This model is ready for inference with TensorRT-LLM in Eagle speculative decoding mode. Eagle modules are used to predict candidate tokens beyond the next token. In the generation step, each forward Eagle module generates a distribution of tokens beyond the previous. Then, a tree-based attention mechanism samples some candidate sequences for the original model to validate. The longest accepted candidate sequence is selected so that more than 1 token is returned in the generation step. The number of tokens generated in each step is called acceptance rate.

Usage

To serve the quantized checkpoint with TensorRT-LLM, follow the sample commands below with the TensorRT-LLM GitHub repo:

trtllm-serve <llama4 checkpoint> --host 0.0.0.0 --port 8000 --backend pytorch --max_batch_size 32 --max_num_tokens 8192 --max_seq_len 8192 --tp_size 8 --extra_llm_api_options extra-llm-api-config.yml

extra-llm-api-config.yml is like this

enable_attention_dp: false
pytorch_backend_config:
  enable_overlap_scheduler: false
  use_cuda_graph: true
  cuda_graph_max_batch_size: 1
  autotuner_enabled: false

speculative_config:
    decoding_type: Eagle
    max_draft_len: 3
    pytorch_eagle_weights_path: <eagle3 checkpoint>

kv_cache_config:
    enable_block_reuse: false

Evaluation

The Eagle acceptance rate benchmark results (MT-Bench) with draft length 3 are presented in the table below:

Category	MT Bench Acceptance Rate
writing	2
roleplay	1.93
reasoning	2.03
math	2.5
coding	2.27
extraction	1.93
stem	2.25
humanities	2.06

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 internal 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.

SUBCARDS:

Explainability

Field:	Response:
Intended Application(s) & Domain(s):	Text generation, reasoning, summarization, and question answering.
Model Type:	Text and Image-to-text transformer
Intended Users:	This model is intended for developers, researchers, and customers building/utilizing LLMs, while balancing accuracy and efficiency.
Output:	Text String(s)
Describe how the model works:	Generates text by predicting the next word or token based on the context provided in the input sequence using multiple self-attention layers
Technical Limitations:	This model is a new technology, and like any new technology, there are risks associated with its use. Testing conducted to date has not covered, nor could it cover, all scenarios. For these reasons, as with all LLMs, this model’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate or other objectionable responses to user prompts. Therefore, before deploying any applications of this model, developers should perform safety testing and tuning tailored to their specific applications of the model. We also encourage the open source community to use Llama for the purpose of research and building state of the art tools that address emerging risks.
Verified to have met prescribed quality standards?	Yes
Performance Metrics:	Accuracy, Throughput, and user-side throughput
Potential Known Risk	None Known
Licensing:	Your usage is governed by the following license Built with Llama.

Bias

Field:	Response:
Participation considerations from adversely impacted groups (protected classes) in model design and testing:	None
Measures taken to mitigate against unwanted bias:	None

Safety & Security

Field:	Response:
Model Application(s):	Chat, Instruction Following, Chatbot Development, Code Generation, Reasoning
Describe life critical application (if present):	None Known
Use Case Restrictions:	Abide by the license Built with Llama.
Model and Dataset Restrictions:	The Principle of least privilege (PoLP) is applied limiting access for dataset generation. Restrictions enforce dataset access during training, and dataset license constraints adhered to. Model checkpoints are made available on Hugging Face, and may become available on cloud providers' model catalog.

Privacy

Field:	Response:
Generatable or Reverse engineerable personal data?	None
Was consent obtained for any personal data used?	None Known
Personal data used to create this model?	None Known
How often is dataset reviewed?	Before Release
Is there provenance for all datasets used in training?	Yes
Does data labeling (annotation, metadata) comply with privacy laws?	Yes
Applicable NVIDIA Privacy Policy	https://www.nvidia.com/en-us/about-nvidia/privacy-policy/