NVIDIA Llama-3_1-Nemotron-Ultra-253B-v1-GGUF Open-Source Model - Free Deployment Compatible with Multiple Hardware Environments

Nvidia Llama 3 1 Nemotron Ultra 253B V1 GGUF

Developed by bartowski

This is the quantized version of the NVIDIA Llama-3_1-Nemotron-Ultra-253B-v1 model, quantized using llama.cpp, supporting multiple quantization types and suitable for various hardware environments.

Large Language Model EnglishOpen Source License:Other #Ultra-large scale parameters #Multi-turn dialogue optimization #High-precision quantization

Downloads 1,607

Release Time : 4/8/2025

Model Overview

A quantized version based on the NVIDIA Llama-3_1-Nemotron-Ultra-253B-v1 model, optimized via the llama.cpp tool, offering multiple quantization options to accommodate different computational resource needs.

Model Features

Multiple quantization options

Offers various quantization types from Q8_0 to IQ2_M, catering to different performance and storage requirements.

High-performance inference

The optimized model significantly reduces computational resource demands while maintaining high-quality output.

Broad compatibility

Supports running in LM Studio, llama.cpp, and projects based on llama.cpp.

Model Capabilities

Text generation

Natural language processing

Dialogue systems

Use Cases

Text generation

Dialogue systems

Used to build intelligent conversational assistants, providing natural and smooth interaction experiences.

Content creation

Assists in generating creative content such as articles, stories, and poems.

Research and development

Model optimization research

Used for researching quantization techniques and performance optimization of large language models.

🚀 Llamacpp imatrix Quantizations of Llama-3_1-Nemotron-Ultra-253B-v1 by nvidia

This project provides Llama-3_1-Nemotron-Ultra-253B-v1 quantizations using llama.cpp. It offers various quantization types for different performance and quality requirements, and provides guidance on downloading and using these quantized models.

🚀 Quick Start

Model Information

Property	Details
Quantized By	bartowski
Pipeline Tag	text-generation
Base Model	nvidia/Llama-3_1-Nemotron-Ultra-253B-v1
License	other
License Name	nvidia-open-model-license
License Link	https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/
Tags	nvidia, llama-3
Language	en
Base Model Relation	quantized

Quantization Method

Using llama.cpp release b5270 for quantization. All quants are made using the imatrix option with the dataset from here.

Running the Model

LM Studio: You can run the quantized models in LM Studio.
llama.cpp: Run them directly with llama.cpp, or any other llama.cpp based project.

✨ Features

Prompt Format

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>

{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

Downloadable Files

Filename	Quant type	File Size	Split	Description
Llama-3_1-Nemotron-Ultra-253B-v1-Q8_0.gguf	Q8_0	269.25GB	true	Extremely high quality, generally unneeded but max available quant.
Llama-3_1-Nemotron-Ultra-253B-v1-Q6_K.gguf	Q6_K	207.88GB	true	Very high quality, near perfect, recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-Q5_K_M.gguf	Q5_K_M	178.55GB	true	High quality, recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-Q5_K_S.gguf	Q5_K_S	174.51GB	true	High quality, recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-Q4_1.gguf	Q4_1	158.80GB	true	Legacy format, similar performance to Q4_K_S but with improved tokens/watt on Apple silicon.
Llama-3_1-Nemotron-Ultra-253B-v1-Q4_K_L.gguf	Q4_K_L	152.50GB	true	Uses Q8_0 for embed and output weights. Good quality, recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-Q4_K_M.gguf	Q4_K_M	150.94GB	true	Good quality, default size for most use cases, recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-Q4_K_S.gguf	Q4_K_S	144.38GB	true	Slightly lower quality with more space savings, recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-Q4_0.gguf	Q4_0	143.74GB	true	Legacy format, offers online repacking for ARM and AVX CPU inference.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ4_NL.gguf	IQ4_NL	143.23GB	true	Similar to IQ4_XS, but slightly larger. Offers online repacking for ARM CPU inference.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ4_XS.gguf	IQ4_XS	135.41GB	true	Decent quality, smaller than Q4_K_S with similar performance, recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-Q3_K_XL.gguf	Q3_K_XL	134.54GB	true	Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
Llama-3_1-Nemotron-Ultra-253B-v1-Q3_K_L.gguf	Q3_K_L	132.70GB	true	Lower quality but usable, good for low RAM availability.
Llama-3_1-Nemotron-Ultra-253B-v1-Q3_K_M.gguf	Q3_K_M	121.88GB	true	Low quality.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ3_M.gguf	IQ3_M	113.50GB	true	Medium-low quality, new method with decent performance comparable to Q3_K_M.
Llama-3_1-Nemotron-Ultra-253B-v1-Q3_K_S.gguf	Q3_K_S	109.72GB	true	Low quality, not recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ3_XS.gguf	IQ3_XS	103.32GB	true	Lower quality, new method with decent performance, slightly better than Q3_K_S.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ3_XXS.gguf	IQ3_XXS	97.62GB	true	Lower quality, new method with decent performance, comparable to Q3 quants.
Llama-3_1-Nemotron-Ultra-253B-v1-Q2_K_L.gguf	Q2_K_L	95.45GB	true	Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
Llama-3_1-Nemotron-Ultra-253B-v1-Q2_K.gguf	Q2_K	93.40GB	true	Very low quality but surprisingly usable.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ2_M.gguf	IQ2_M	85.44GB	true	Relatively low quality, uses SOTA techniques to be surprisingly usable.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ2_S.gguf	IQ2_S	78.55GB	true	Low quality, uses SOTA techniques to be usable.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ2_XS.gguf	IQ2_XS	74.88GB	true	Low quality, uses SOTA techniques to be usable.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ2_XXS.gguf	IQ2_XXS	67.44GB	true	Very low quality, uses SOTA techniques to be usable.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ1_M.gguf	IQ1_M	58.82GB	true	Extremely low quality, not recommended.
Llama-3_1-Nemotron-Ultra-253B-v1-IQ1_S.gguf	IQ1_S	53.65GB	true	Extremely low quality, not recommended.

Embed/Output Weights

Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.

📦 Installation

Downloading using huggingface-cli

Click to view download instructions

First, make sure you have hugginface-cli installed:

pip install -U "huggingface_hub[cli]"

Then, you can target the specific file you want:

huggingface-cli download bartowski/nvidia_Llama-3_1-Nemotron-Ultra-253B-v1-GGUF --include "nvidia_Llama-3_1-Nemotron-Ultra-253B-v1-Q4_K_M.gguf" --local-dir ./

If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:

huggingface-cli download bartowski/nvidia_Llama-3_1-Nemotron-Ultra-253B-v1-GGUF --include "nvidia_Llama-3_1-Nemotron-Ultra-253B-v1-Q8_0/*" --local-dir ./

You can either specify a new local-dir (nvidia_Llama-3_1-Nemotron-Ultra-253B-v1-Q8_0) or download them all in place (./)

🔧 Technical Details

ARM/AVX Information

Previously, you would download Q4_0_4_4/4_8/8_8, and these would have their weights interleaved in memory in order to improve performance on ARM and AVX machines by loading up more data in one pass.

Now, however, there is something called "online repacking" for weights. details in this PR. If you use Q4_0 and your hardware would benefit from repacking weights, it will do it automatically on the fly.

As of llama.cpp build b4282 you will not be able to run the Q4_0_X_X files and will instead need to use Q4_0.

Additionally, if you want to get slightly better quality for , you can use IQ4_NL thanks to this PR which will also repack the weights for ARM, though only the 4_4 for now. The loading time may be slower but it will result in an overall speed incrase.

Click to view Q4_0_X_X information (deprecated

I'm keeping this section to show the potential theoretical uplift in performance from using the Q4_0 with online repacking.

Click to view benchmarks on an AVX2 system (EPYC7702)

model	size	params	backend	threads	test	t/s	% (vs Q4_0)
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	pp512	204.03 ± 1.03	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	pp1024	282.92 ± 0.19	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	pp2048	259.49 ± 0.44	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	tg128	39.12 ± 0.27	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	tg256	39.31 ± 0.69	100%
qwen2 3B Q4_0	1.70 GiB	3.09 B	CPU	64	tg512	40.52 ± 0.03	100%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	pp512	301.02 ± 1.74	147%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	pp1024	287.23 ± 0.20	101%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	pp2048	262.77 ± 1.81	101%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	tg128	18.80 ± 0.99	48%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	tg256	24.46 ± 3.04	83%
qwen2 3B Q4_K_M	1.79 GiB	3.09 B	CPU	64	tg512	36.32 ± 3.59	90%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	pp512	271.71 ± 3.53	133%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	pp1024	279.86 ± 45.63	100%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	pp2048	320.77 ± 5.00	124%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	tg128	43.51 ± 0.05	111%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	tg256	43.35 ± 0.09	110%
qwen2 3B Q4_0_8_8	1.69 GiB	3.09 B	CPU	64	tg512	42.60 ± 0.31	105%

Q4_0_8_8 offers a nice bump to prompt processing and a small bump to text generation

Which File to Choose

Click here for details

A great write up with charts showing various performances is provided by Artefact2 here

The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.

If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.

If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.

Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.

If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.

If you want to get more into the weeds, you can check out this extremely useful feature chart:

llama.cpp feature matrix

But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.

These I-quants can also be used on CPU, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.

📄 License

This project is licensed under the nvidia-open-model-license.

Credits

Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset.

Thank you ZeroWw for the inspiration to experiment with embed/output.

Thank you to LM Studio for sponsoring my work.

Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご