Model Overview

Description:

The NVIDIA Nemotron-3.5-Lightning-30B-A3B-NVFP4-DFlash model is the DFlash speculative decoding checkpoint for NVIDIA's Nemotron-3.5-Lightning-30B-A3B model family, which is a hybrid LatentMoE language model designed for reasoning, chat, and agentic workflows. For more information, please check BF16, NVFP4. The NVIDIA Nemotron-3.5-Lightning-30B-A3B-DFlash-NVFP4 model is intended for lower-latency speculative decoding deployments tuned for low-concurrency data centre and workstation workflows.

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

License/Terms of Use:

GOVERNING DOWNLOAD TERMS: Use of this model is governed by the OpenMDW-1.1 model license.

Deployment Geography:

Global

Use Case:

Developers deploying Nemotron-3.5-Lightning-30B-A3B for reasoning, chat, RAG, and agentic workflows that benefit from lower-latency speculative decoding on data centre GPUs and high-end local GPU systems. This release is intended for DFlash-assisted serving of Nemotron-3.5-Lightning-30B-A3B rather than as a standalone target model checkpoint.

Release Date:

Hugging Face 08/11/2026 via https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DFlash

References

• NVIDIA Model Optimizer: https://github.com/NVIDIA/Model-Optimizer
• NVIDIA Nemotron-3.5-Lightning-30B-A3B-BF16 reasoning model card
• NVIDIA Nemotron-3.5-Lightning-30B-A3B-NVFP4 reasoning model card
• OpenMDW License Agreement, version 1.1
• DFlash: Block Diffusion for Flash Speculative Decoding

Model Architecture:

The DFlash model architecture is as follows:

Architecture Type: Dense GQA (Dense MLP + GQA Attention)
Network Architecture: Dense FFN MLP, and GQA Attention layers; DFlash speculative decoding attention uses non-causal, full-sequence grouped-query attention (GQA). Number of Model Parameters: 833M total parameters, of which 481M are non-embedding parameters.

For more information about the underlying model's architecture, please see this Nemotron-3.5-Lightning-30B-A3B-BF16, Nemotron-3.5-Lightning-30B-A3B-NVFP4.

Input:

Input Type(s): Text
Input Format(s): String
Input Parameters: One-Dimensional (1D): Sequences
Other Properties Related to Input: Maximum context length up to 1M tokens. Supported languages include English, Spanish, French, German, Italian, and Japanese.

Output:

Output Type(s): Text
Output Format: String
Output Parameters: One-Dimensional (1D): Sequences
Other Properties Related to Output: Outputs may include natural-language responses, reasoning traces, tool-use content, and structured outputs depending on chat-template configuration and application-level tooling.

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):

• vLLM
• llama.cpp

Supported Hardware Microarchitecture Compatibility:

• NVIDIA Blackwell GB200

Preferred Operating System(s):

• Linux

The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.

Model Version(s):

The model is a Nemotron-3.5-Lightning-30B-A3B-NVFP4-DFlash speculative decoding version of the model and quantized with Model Optimizer 0.45.0 version

Training and Evaluation Datasets:

Training Dataset:

For more information about the underlying Nemotron-3.5-Lightning-30B-A3B model, please visit the model card - Nemotron-3.5-Lightning-30B-A3B-NVFP4, Nemotron-3.5-Lightning-30B-A3B-BF16.

Link: Nemotron-Post-Training-Dataset-v2, and Nemotron-Post-Training-Dataset-v3 only prompts from the datasets were used for data synthesis, (the original responses from GPT were not used), which is then used to train the DFlash modules.
Data Modality: Text
Text Training Data Size: 66 Billion Tokens, repeated for 2 epochs
Data Collection Method by dataset: Hybrid: Automated, manually-collected, Synthetic
Labeling Method by dataset: Hybrid: Automated, manually-labelled, Synthetic
Properties: The Nemotron-3.5-Lightning-30B-A3B-NVFP4-DFlash model was trained exclusively on a post-training corpus includes synthetic data for reasoning, code, science, tool use, instruction following, structured outputs, and multilingual tasks. Dataset disclosures are published in the nvidia/nemotron-post-training-dataset-v2 and nvidia/nemotron-post-training-v3 Hugging Face collections.

For more information about the datasets used to train this model, please see the Public Summary of Training Content

Evaluation Dataset:

The model is evaluated on the SPEED-Bench dataset for speculative decoding benchmarking.

Data Collection Method by dataset: Hybrid: Automated, manually-collected, Synthetic
Labeling Method by dataset: Hybrid: Automated, manually-labelled, Synthetic
Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for LLMs. These benchmarks prompt the underlying model on a diverse set of tasks such as coding, math, writing, translation, etc., and the acceptance rates achieved from speculative decoding over the responses are measured in order to evaluate the quality of the speculation model.

Inference:

Acceleration Engine: vLLM, llama.cpp
Test Hardware: NVIDIA Hopper - H100; NVIDIA Blackwell - GB200; NVIDIA Blackwell - GeForce RTX 5090

DFlash Speculative Decoding

Nemotron-3.5-Lightning-30B-A3B supports multiple speculative decoding paths, including native MTP, DFlash, and DSpark for DGX Spark and low-concurrency data centre workflows. This release is the DFlash speculative decoding checkpoint intended to be paired with the Nemotron-3.5-Lightning-30B-A3B target model in vLLM or llama.cpp. DFlash is designed to reduce end-to-end latency for small-batch or single-request serving by drafting more than one candidate token per target-model verification step.

Usage

To serve the checkpoint with vLLM, refer to the DFlash recipes in the NVIDIA Nemotron-3.5-Lightning-30B-A3B Model Card.

For local workflows, the Model Card also includes recipes for deployment with llama.cpp on NVIDIA Blackwell - GeForce RTX 5090.

Evaluation

Acceptance rate on SPEED-Bench with draft length 7:

Baseline: NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 Benchmarked with temperature=1.0, top_p=0.95.

Model Limitations:

The base model may generate inaccurate, incomplete, or otherwise undesirable responses, even when prompts are benign. It may reflect biases or content artifacts present in its training data, and output quality can vary by domain, language, reasoning depth, and context length. Developers should add application-specific safeguards and evaluate the system in the intended deployment environment before production use.

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. 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++ Bias, Explainability, Safety & Security, and Privacy Subcards.

Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.

SUBCARDS:

Explainability

Bias

Safety & Security

Privacy