DeepSeek-R1-Distill-Llama-8B-NexaQuant

Background + Overview

DeepSeek-R1 has been making headlines for rivaling OpenAI’s O1 reasoning model while remaining fully open-source. Many users want to run it locally to ensure data privacy, reduce latency, and maintain offline access. However, fitting such a large model onto personal devices typically requires quantization (e.g. Q4_K_M), which often sacrifices accuracy (up to ~22% accuracy loss) and undermines the benefits of the local reasoning model.

We’ve solved the trade-off by quantizing the DeepSeek R1 Distilled model to one-fourth its original size—without losing any accuracy. This lets you run powerful on-device reasoning wherever you are, with no compromises. Tests on an HP Omnibook AIPC with an AMD Ryzen™ AI 9 HX 370 processor showed a decoding speed of 17.20 tokens per second and a peak RAM usage of just 5017 MB in NexaQuant version—compared to only 5.30 tokens per second and 15564 MB RAM in the unquantized version—while maintaining full precision model accuracy.

How to run locally

NexaQuant is compatible with Nexa-SDK, Ollama, LM Studio, Llama.cpp, and any llama.cpp based project. Below, we outline multiple ways to run the model locally.

Option 1: Using Nexa SDK

Step 1: Install Nexa SDK

Follow the installation instructions in Nexa SDK's GitHub repository.

Step 2: Run the model with Nexa

Execute the following command in your terminal:

nexa run DeepSeek-R1-Distill-Llama-8B-NexaQuant:q4_0

Option 2: Using llama.cpp

Step 1: Build llama.cpp on Your Device

Follow the "Building the project" instructions in the llama.cpp repository to build the project.

Step 2: Run the Model with llama.cpp

Once built, run llama-cli under <build_dir>/bin/:

./llama-cli \
    --model your/local/path/to/DeepSeek-R1-Distill-Llama-8B-NexaQuant \
    --prompt 'Provide step-by-step reasoning enclosed in <think> </think> tags, followed by the final answer enclosed in \boxed{} tags.' \

Option 3: Using LM Studio

Step 1: Download and Install LM Studio

Get the latest version from the official website.

Step 2: Load and Run the Model

In LM Studio's top panel, search for and select NexaAIDev/DeepSeek-R1-Distill-Llama-8B-NexaQuant.
Click Download (if not already downloaded) and wait for the model to load.
Once loaded, go to the chat window and start a conversation.

Example

On the left, we have an example of what LMStudio Q4_K_M responded. On the right is our NexaQuant version.

Prompt: A Common Investment Banking BrainTeaser Question

There is a 6x8 rectangular chocolate bar made up of small 1x1 bits. We want to break it into the 48 bits. We can break one piece of chocolate horizontally or vertically, but cannot break two pieces together! What is the minimum number of breaks required?

Right Answer: 47

Benchmarks

NexaQuant on Reasoning Benchmarks Compared to BF16 and LMStudio's Q4_K_M

8B:

The general capacity has also greatly improved:

8B:

Benchmark	Full 16-bit	llama.cpp (4-bit)	NexaQuant (4-bit)
HellaSwag	57.07	52.12	54.56
MMLU	55.59	52.82	54.94
Humanities	50.48	44.74	50.24
Social Sciences	65.32	65.62	64.74
STEM	47.73	52.50	46.75
ARC Easy	74.49	69.32	71.72
MathQA	35.34	30.00	32.46
PIQA	78.56	76.09	77.68
IFEval - Inst - Loose	44.24	44.95	42.45
IFEval - Inst - Strict	42.57	44.95	40.05
IFEval - Prom - Loose	30.31	25.74	28.47
IFEval - Prom - Strict	27.91	25.74	25.51

What's next

Inference Nexa Quantized Deepseek-R1 distilled model on NPU.
This model is designed for complex problem-solving, which is why it has a longer thinking process. We understand this can be an issue in some cases, and we're actively working on improvements.

If you liked our work, feel free to ⭐Star Nexa's GitHub Repo.

Interested in running DeepSeek R1 on your own devices with optimized CPU, GPU, and NPU acceleration or compressing your finetuned DeepSeek-Distill-R1? Let’s chat!

Blogs | Discord | X(Twitter)

NexaAI
/

DeepSeek-R1-Distill-Llama-8B-NexaQuant