nvidia/DeepSeek-V4-Flash-0731-NVFP4

Model Overview

Description:

The NVIDIA DeepSeek-V4-Flash-0731-NVFP4 model is the quantized version of DeepSeek AI's DeepSeek-V4-Flash-0731 model, an autoregressive Mixture-of-Experts language model that uses an optimized Transformer architecture with hybrid attention (Compressed Sparse Attention and Heavily Compressed Attention) and Manifold-Constrained Hyper-Connections. DeepSeek-V4-Flash-0731 is the official DeepSeek-V4-Flash release and includes DeepSeek's DSpark speculative decoding module in the same checkpoint. For more information, refer to the DeepSeek-V4-Flash-0731 model card. The NVIDIA DeepSeek-V4-Flash-0731-NVFP4 model is quantized with Model Optimizer.

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

Third-Party Community Consideration

This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party’s requirements for this application and use case; see link to Non-NVIDIA (DeepSeek-V4-Flash-0731) Model Card.

License/Terms of Use:

MIT

Deployment Geography:

Global

Use Case:

DeepSeek V4 is well-suited for advanced reasoning, agentic AI applications, tool use scenarios, and complex problem-solving in domains such as mathematics, software engineering, and enterprise AI assistants.

Release Date:

Hugging Face 08/31/2026 via https://huggingface.co/nvidia/DeepSeek-V4-Flash-0731-NVFP4

References

NVIDIA Model Optimizer: https://github.com/NVIDIA/Model-Optimizer

Model Architecture:

Architecture Type: Transformers
Network Architecture: DeepseekV4ForCausalLM — Mixture-of-Experts (MoE) with Hybrid Attention (Compressed Sparse Attention + Heavily Compressed Attention)
Number of Model Parameters: 304B in total and 13B activated
This model was developed based on DeepSeek-V4-Flash-0731

Input:

Input Type(s): Text
Input Format(s): String
Input Parameters: One-Dimensional (1D): Sequences
Other Properties Related to Input: Supports multi-turn conversations with system prompts, user messages, and assistant responses. Maximum context length of 1 million tokens. Uses a custom encoding pipeline (encoding_dsv4) with reasoning effort levels (low, high, max).

Output:

Output Type(s): Text
Output Format: String
Output Parameters: One-Dimensional (1D): Sequences
Other Properties Related to Output: Supports structured JSON output, function/tool calling, and reasoning content when enabled.

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):

• SGLang
  
• vLLM
  

Supported Hardware Microarchitecture Compatibility:

• NVIDIA Blackwell
  

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 DeepSeek-V4-Flash-0731 NVFP4 quantized with nvidia-modelopt v0.46.0

Training and Evaluation Datasets:

Calibration Dataset:

Link: cnn_dailymail, Nemotron-Post-Training-Dataset-v2
Data Collection Method by dataset: Automated.
Labeling Method by dataset: Automated.
Properties: The cnn_dailymail dataset is an English-language dataset containing just over 300k unique news articles as written by journalists at CNN and the Daily Mail. The Nemotron-Post-Training-Dataset-v2 is a post-training dataset curated by NVIDIA containing multi-turn conversations across diverse topics.

Training Dataset:

Data Modality: Undisclosed
Data Collection Method by dataset: Undisclosed
Labeling Method by dataset: Undisclosed
Properties: Undisclosed

Evaluation Dataset:

Datasets: GPQA Diamond, AA-LCR, τ²-Bench Telecom, SciCode, IFBench, Terminal-Bench v2.1
Data Collection Method by dataset: Hybrid: Automated, Manually-Collected
Labeling Method by dataset: Hybrid: Manually-Labeled, Automated
Properties: We evaluated the model on reasoning, long-context recall, agentic tool-use, coding, and instruction-following benchmarks: GPQA Diamond contains 448 graduate-level multiple-choice questions written by domain experts in biology, physics, and chemistry. AA-LCR (Artificial Analysis Long Context Recall) evaluates a model's ability to accurately retrieve and recall information from long input contexts. τ²-Bench Telecom evaluates agentic tool-use and policy-adherence capabilities in dual-control telecom customer-service scenarios where the model interacts with a simulated user and external tools to resolve account issues. SciCode evaluates scientific coding capabilities. IFBench is a benchmark for evaluating instruction-following capabilities across diverse and structured task constraints. Terminal-Bench v2.1 evaluates agentic software-engineering capability in a real terminal environment, where the model must plan and execute shell commands to complete tasks end to end.

Inference:

Acceleration Engine: SGLang and vLLM
Test Hardware: NVIDIA B200

Post Training Quantization

This model was obtained by quantizing the weights and activations of DeepSeek-V4-Flash-0731 to NVFP4 data type, ready for inference with SGLang and vLLM. Only the routed MoE experts are quantized. DeepSeek's DSpark speculative-decoding heads are carried through unquantized and unmodified.

Quantization Recipe

The recipe is in the Model Optimizer repository under examples/deepseek/deepseek_v4 (ptq.py, quantize_to_nvfp4.py). It runs in three stages:

1. Reshard the source checkpoint to the model-parallel layout (CPU only), using DeepSeek's own convert.py:

python <DeepSeek-V4-Flash-0731>/inference/convert.py \
    --hf-ckpt-path <DeepSeek-V4-Flash-0731> \
    --save-path <mp8_checkpoint> \
    --n-experts 256 \
    --model-parallel 8

2. Collect activation ranges (amax) on the routed experts (from Model Optimizer repo root):

torchrun --nproc-per-node 8 examples/deepseek/deepseek_v4/ptq.py \
    --model_path <mp8_checkpoint> \
    --config <DeepSeek-V4-Flash-0731>/inference/config.json \
    --dsv4_inference_dir <DeepSeek-V4-Flash-0731>/inference \
    --output_path <amax_dump>

Calibration uses the script defaults: cnn_dailymail and Nemotron-Post-Training-Dataset-v2, --calib_size 64, --calib_seq 512.

3. Export the NVFP4 checkpoint (from Model Optimizer repo root):

python examples/deepseek/deepseek_v4/quantize_to_nvfp4.py \
    --amax_path <amax_dump> \
    --source_ckpt <DeepSeek-V4-Flash-0731> \
    --output_ckpt <output> \
    --cast_mxfp4_to_nvfp4

--cast_mxfp4_to_nvfp4 matters for this model in particular. DeepSeek-V4 already ships its routed experts in MXFP4 (E2M1 nibbles with one E8M0 scale per 32 elements). NVFP4 reuses the same E2M1 nibble grid but stores an E4M3 scale per 16 elements, so the weight bits are carried over unchanged — a lossless bit-cast — and only the block scales are rewritten. The activation amax from stage 2 is therefore the only calibrated quantity that survives into the artifact, and the checkpoint is slightly larger than its source (scale bytes double while weight bytes stay the same).

Usage

Deploy with SGLang

sglang serve \
    --model-path nvidia/DeepSeek-V4-Flash-0731-NVFP4 \
    --tp-size 8 \
    --kv-cache-dtype fp8_e4m3 \
    --moe-runner-backend flashinfer_trtllm_routed \
    --chunked-prefill-size 4096 \
    --swa-full-tokens-ratio 0.1 \
    --tool-call-parser deepseekv4 \
    --reasoning-parser deepseek-v4 \
    --trust-remote-code

Deploy with vLLM

vllm serve nvidia/DeepSeek-V4-Flash-0731-NVFP4 \
    --tensor-parallel-size 8 \
    --enable-expert-parallel \
    --max-model-len 393216 \
    --kv-cache-dtype fp8 \
    --block-size 256 \
    --attention_config.use_fp4_indexer_cache=True \
    --tokenizer-mode deepseek_v4 \
    --tool-call-parser deepseek_v4 \
    --enable-auto-tool-choice \
    --reasoning-parser deepseek_v4 \
    --reasoning-config '{"reasoning_parser":"deepseek_v4","reasoning_start_str":"","reasoning_end_str":""}' \
    --trust-remote-code

The base DeepSeek-V4-Flash-0731 checkpoint includes DSpark and those heads are preserved here, but speculative decoding was not exercised during this release's validation — enable it only after verifying it for this NVFP4 checkpoint.

Evaluation

The accuracy benchmark results are presented in the table below:

Precision	GPQA Diamond	AA-LCR	τ²-Bench Telecom	SciCode	IFBench	Terminal-Bench v2.1
MXFP4 (source)	91.5	72.1	98.7	51.7	75.8	74.7
NVFP4	91.5	71.8	97.9	52.1	75.5	73.7

Baseline: DeepSeek-V4-Flash-0731. Measured at temperature=1.0, top_p=1.0, max reasoning effort, max_new_tokens=384000; τ²-Bench Telecom and Terminal-Bench v2.1 use top_p=0.95.

Model Limitations:

The base model was trained on data that contains toxic language and societal biases originally crawled from the internet. Therefore, the model may amplify those biases and return toxic responses especially when prompted with toxic prompts. The model may generate answers that may be inaccurate, omit key information, or include irrelevant or redundant text producing socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive.

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.

Users are responsible for model inputs and outputs. Users are responsible for ensuring safe integration of this model, including implementing guardrails as well as other safety mechanisms, prior to deployment.

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