🌠Qwen3-Next: Run Locally Guide

Run Qwen3-Next-80B-A3B-Instruct and Thinking versions locally on your device!

Qwen released Qwen3-Next in Sept 2025, which are 80B MoEs with Thinking and Instruct model variants of Qwen3. With 256K context, Qwen3-Next was designed with a brand new architecture (Hybrid of MoEs & Gated DeltaNet + Gated Attention) that specifically optimizes for fast inference on longer context lengths. Qwen3-Next has 10x faster inference than Qwen3-32B.

Run Qwen3-Next InstructRun Qwen3-Next Thinking

Qwen3-Next-80B-A3B Dynamic GGUFs: Instruct • Thinking

⚙️Usage Guide

The thinking model uses temperature = 0.6, but the instruct model uses temperature = 0.7 The thinking model uses top_p = 0.95, but the instruct model uses top_p = 0.8

To achieve optimal performance, Qwen recommends these settings:

Instruct:	Thinking:
Temperature = 0.7	Temperature = 0.6
Min_P = 0.00 (llama.cpp's default is 0.1)	Min_P = 0.00 (llama.cpp's default is 0.1)
Top_P = 0.80	Top_P = 0.95
TopK = 20	TopK = 20
presence_penalty = 0.0 to 2.0 (llama.cpp default turns it off, but to reduce repetitions, you can use this)	presence_penalty = 0.0 to 2.0 (llama.cpp default turns it off, but to reduce repetitions, you can use this)

Adequate Output Length: Use an output length of 32,768 tokens for most queries for the thinking variant, and 16,384 for the instruct variant. You can increase the max output size for the thinking model if necessary.

Chat template for both Thinking (thinking has <think></think>) and Instruct is below:

<|im_start|>user
Hey there!<|im_end|>
<|im_start|>assistant
What is 1+1?<|im_end|>
<|im_start|>user
2<|im_end|>
<|im_start|>assistant

📖 Run Qwen3-Next Tutorials

Below are guides for the Thinking and Instruct versions of the model.

Instruct: Qwen3-Next-80B-A3B-Instruct

Given that this is a non thinking model, there is no need to set thinking=False and the model does not generate <think> </think> blocks.

⚙️Best Practices

To achieve optimal performance, Qwen recommends the following settings:

• We suggest using temperature=0.7, top_p=0.8, top_k=20, and min_p=0.0 presence_penalty between 0 and 2 if the framework supports to reduce endless repetitions.

• temperature = 0.7

• top_k = 20

• min_p = 0.00 (llama.cpp's default is 0.1)

• top_p = 0.80

• presence_penalty = 0.0 to 2.0 (llama.cpp default turns it off, but to reduce repetitions, you can use this) Try 1.0 for example.

• Supports up to 262,144 context natively but you can set it to 32,768 tokens for less RAM use

✨ Llama.cpp: Run Qwen3-Next-80B-A3B-Instruct Tutorial

1. Obtain the latest llama.cpp on GitHub here. You can follow the build instructions below as well. Change -DGGML_CUDA=ON to -DGGML_CUDA=OFF if you don't have a GPU or just want CPU inference.

apt-get update
apt-get install pciutils build-essential cmake curl libcurl4-openssl-dev -y
git clone https://github.com/ggml-org/llama.cpp
cmake llama.cpp -B llama.cpp/build \
    -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=ON -DLLAMA_CURL=ON
cmake --build llama.cpp/build --config Release -j --clean-first --target llama-cli llama-gguf-split
cp llama.cpp/build/bin/llama-* llama.cpp

1. You can directly pull from HuggingFace via:

   ./llama.cpp/llama-cli \
       -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:Q4_K_XL \
       --jinja -ngl 99 --threads -1 --ctx-size 32684 \
       --temp 0.7 --min-p 0.0 --top-p 0.80 --top-k 20 --presence-penalty 1.0

2. Download the model via (after installing pip install huggingface_hub hf_transfer ). You can choose UD_Q4_K_XL or other quantized versions.

# !pip install huggingface_hub hf_transfer
import os
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
from huggingface_hub import snapshot_download
snapshot_download(
    repo_id = "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF",
    local_dir = "Qwen3-Next-80B-A3B-Instruct-GGUF",
    allow_patterns = ["*UD-Q4_K_XL*"],
)

Thinking: Qwen3-Next-80B-A3B-Thinking

This model supports only thinking mode and a 256K context window natively. The default chat template adds <think> automatically, so you may see only a closing </think> tag in the output.

⚙️Best Practices

To achieve optimal performance, Qwen recommends the following settings:

• We suggest using temperature=0.6, top_p=0.95, top_k=20, and min_p=0.0 presence_penalty between 0 and 2 if the framework supports to reduce endless repetitions.

• temperature = 0.6

• top_k = 20

• min_p = 0.00 (llama.cpp's default is 0.1)

• top_p = 0.95

• presence_penalty = 0.0 to 2.0 (llama.cpp default turns it off, but to reduce repetitions, you can use this) Try 1.0 for example.

• Supports up to 262,144 context natively but you can set it to 32,768 tokens for less RAM use

✨ Llama.cpp: Run Qwen3-Next-80B-A3B-Thinking Tutorial

1. Obtain the latest llama.cpp on GitHub here. You can follow the build instructions below as well. Change -DGGML_CUDA=ON to -DGGML_CUDA=OFF if you don't have a GPU or just want CPU inference.

apt-get update
apt-get install pciutils build-essential cmake curl libcurl4-openssl-dev -y
git clone https://github.com/ggml-org/llama.cpp
cmake llama.cpp -B llama.cpp/build \
    -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=ON -DLLAMA_CURL=ON
cmake --build llama.cpp/build --config Release -j --clean-first --target llama-cli llama-gguf-split
cp llama.cpp/build/bin/llama-* llama.cpp

1. You can directly pull from Hugging Face via:

   ./llama.cpp/llama-cli \
       -hf unsloth/Qwen3-Next-80B-A3B-Thinking-GGUF:Q4_K_XL \
       --jinja -ngl 99 --threads -1 --ctx-size 32684 \
       --temp 0.6 --min-p 0.0 --top-p 0.95 --top-k 20 --presence-penalty 1.0

2. Download the model via (after installing pip install huggingface_hub hf_transfer ). You can choose UD_Q4_K_XL or other quantized versions.

🛠️ Improving generation speed

If you have more VRAM, you can try offloading more MoE layers, or offloading whole layers themselves.

Normally, -ot ".ffn_.*_exps.=CPU" offloads all MoE layers to the CPU! This effectively allows you to fit all non MoE layers on 1 GPU, improving generation speeds. You can customize the regex expression to fit more layers if you have more GPU capacity.

If you have a bit more GPU memory, try -ot ".ffn_(up|down)_exps.=CPU" This offloads up and down projection MoE layers.

Try -ot ".ffn_(up)_exps.=CPU" if you have even more GPU memory. This offloads only up projection MoE layers.

You can also customize the regex, for example -ot "\.(6|7|8|9|[0-9][0-9]|[0-9][0-9][0-9])\.ffn_(gate|up|down)_exps.=CPU" means to offload gate, up and down MoE layers but only from the 6th layer onwards.

The latest llama.cpp release also introduces high throughput mode. Use llama-parallel. Read more about it here. You can also quantize the KV cache to 4bits for example to reduce VRAM / RAM movement, which can also make the generation process faster. The next section talks about KV cache quantization.

📐How to fit long context

To fit longer context, you can use KV cache quantization to quantize the K and V caches to lower bits. This can also increase generation speed due to reduced RAM / VRAM data movement. The allowed options for K quantization (default is f16) include the below.

--cache-type-k f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1

You should use the _1 variants for somewhat increased accuracy, albeit it's slightly slower. For eg q4_1, q5_1 So try out --cache-type-k q4_1

You can also quantize the V cache, but you will need to compile llama.cpp with Flash Attention support via -DGGML_CUDA_FA_ALL_QUANTS=ON, and use --flash-attn to enable it. After installing Flash Attention, you can then use --cache-type-v q4_1