No official torrent yet — requesting it tells the flock what to host next. You cannot seed a catalog-only page.
Cataloged from nvidia/DeepSeek-V4.1-Flash-NVFP4
All models by nvidia →The NVIDIA DeepSeek-V4.1-Flash NVFP4 model is the quantized version of DeepSeek AI's DeepSeek-V4.1-Flash model, which is an auto-regressive language model that uses an optimized transformer architecture. DeepSeek-V4.1-Flash is a natively multimodal Mixture-of-Experts (MoE) model with a Causal Encoder-Decoder (CED) architecture, Compressed Sparse Attention 2 (CSA2), and support for contexts of up to one million tokens. For more information, please check here. The NVIDIA DeepSeek-V4.1-Flash NVFP4 model is quantized with Model Optimizer.
This model is ready for commercial or non-commercial use.
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-V4.1-Flash) Model Card from DeepSeek AI.
Global
Developers looking to take off-the-shelf, pre-quantized models for deployment in AI Agent systems, chatbots, RAG systems, and other AI-powered applications.
Hugging Face 09/16/2026 via https://huggingface.co/nvidia/DeepSeek-V4.1-Flash-NVFP4
NVIDIA Model Optimizer: https://github.com/NVIDIA/Model-Optimizer
Architecture Type: Transformers
Network Architecture: DeepSeek-V4.1-Flash (DeepseekV41ForCausalLM) — Causal Encoder-Decoder Mixture-of-Experts with Compressed Sparse Attention 2
Number of Model Parameters: 552B backbone in total, 8B activated during prefill and 16B during decode, plus 196B Engram conditional memory
This model was developed based on DeepSeek-V4.1-Flash
Input Type(s): Text, Image
Input Format(s): String, Red, Green, Blue (RGB)
Input Parameters: One-Dimensional (1D), Two-Dimensional (2D)
Other Properties Related to Input: Context length up to 1M
Output Type(s): Text
Output Format: String
Output Parameters: One-Dimensional (1D): Sequences
Other Properties Related to Output: None
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.
Supported Runtime Engine(s):
Both runtimes have been tested with this NVFP4 checkpoint on GB300. The benchmark results below were obtained with vLLM; SGLang was validated through loading, generation, reasoning/tool-call parsing, and image-input smoke tests.
Supported Hardware Microarchitecture Compatibility:
Preferred Operating System(s):
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.
The model is quantized with nvidia-modelopt v0.47.0rc0 and V4.1-specific compatibility changes. Source revision: dba1be0a40aa45a94ad051997016db3960a90277.
** 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.
Calibration used 512 samples per dataset (1,024 total), sequence length 512, batch size 4, and selection seed 0. The Nemotron subsets were stem, chat, math, and code.
** Data Modality: Undisclosed
** Data Collection Method by dataset: Undisclosed
** Labeling Method by dataset: Undisclosed
** Properties: Undisclosed
Acceleration Engines: vLLM and SGLang
Test Hardware: NVIDIA GB300 (Blackwell)
This model converts the ordinary routed MoE experts from source MXFP4 to NVFP4 weights and activations (W4A4), with group size 16. The converted projections are w1, w2, and w3 for 384 experts across 40 layers. Attention, shared experts, vision, Engram lookup tables, MTP/DSpark, and other excluded components retain their source precision, including MXFP8 where applicable.
All 16,986,931,200 weight blocks passed lossless conversion, preserving dequantized weight values with signed-zero canonicalization. Activation scales were calibrated; 18 of 46,080 projection entries used fallback scales. Lossless weight conversion does not imply identical inference outputs.
The source experts already use four-bit weights. The finer NVFP4 scale layout increases checkpoint size from approximately 476 GiB to 492 GiB. The export contains 48 safetensors shards.
Use lmsysorg/sglang:dev-cu13-dsv41 on four GB300 GPUs. The tested build was SGLang commit da64c5cbb8cf6bfd39be19da43573fdfd484c43a. This configuration passed loading, generation, reasoning and tool-call parsing, and image-input smoke tests.
The tested build automatically selects flashinfer_trtllm_routed for the NVFP4 experts. Include a request-level reasoning_effort, such as "max", to enable thinking; thinking is off by default in this SGLang build. During staging, substitute the local checkpoint directory for the repository ID if needed.
Use vllm/vllm-openai:deepseekv41-flash-0909 on four GB300 GPUs. This example adapts the recorded text-serving command to the repository ID. During staging, substitute the local checkpoint directory if needed.
DSpark tensors are preserved, but speculative decoding was not exercised in the reported validation. The vLLM example is text-only; image serving with vLLM requires a multimodal configuration. Validate the final staged package with the intended runtime before deployment.
Accuracy results obtained with vLLM are shown below, in percent.
Baseline: DeepSeek-V4.1-Flash, a mixed-precision checkpoint with MXFP4 routed experts. Ordinary benchmark requests used
temperature=1.0,top_p=0.95,reasoning_effort=100, andmax_new_tokens=262144. GPQA and AA-LCR used 16 repeats; IFBench used 5. Terminal-Bench used a separate agentic configuration with eight trajectories per task.
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.
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.
Please make sure you have proper rights and permissions for all input image and video content; if image or video includes people, personal health information, or intellectual property, the image or video generated will not blur or maintain proportions of image subjects included.
Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.
python -m sglang.launch_server \ --model-path nvidia/DeepSeek-V4.1-Flash-NVFP4 \ --served-model-name DeepSeek-V4.1-Flash-NVFP4 \ --host 127.0.0.1 --port 30000 \ --tp 4 \ --context-length 1048576 \ --reasoning-parser deepseek-v41 \ --tool-call-parser deepseekv41 \ --chunked-prefill-size 4096 \ --max-running-requests 16 \ --log-level infovllm serve nvidia/DeepSeek-V4.1-Flash-NVFP4 \ --tensor-parallel-size 4 \ --data-parallel-size 1 \ --tokenizer-mode deepseek_v41 \ --reasoning-parser deepseek_v41 \ --language-model-only \ --max-model-len 1048576 \ --max-num-seqs 32 \ --max-num-batched-tokens 8192 \ --enable-chunked-prefill \ --no-enable-prefix-caching \ --model-loader-extra-config '{"enable_multithread_load": true, "num_threads": 128}'