Instructions to use unsloth/Qwen3.8-Flash-Next-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use unsloth/Qwen3.8-Flash-Next-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama cli -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Llama new
LM Studio
Jan

vLLM

How to use unsloth/Qwen3.8-Flash-Next-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "unsloth/Qwen3.8-Flash-Next-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "unsloth/Qwen3.8-Flash-Next-GGUF",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Ollama
How to use unsloth/Qwen3.8-Flash-Next-GGUF with Ollama:
```
ollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL
```
Unsloth Desktop

How to use unsloth/Qwen3.8-Flash-Next-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use unsloth/Qwen3.8-Flash-Next-GGUF with Docker Model Runner:
```
docker model run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL
```

Lemonade

How to use unsloth/Qwen3.8-Flash-Next-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.Qwen3.8-Flash-Next-GGUF-UD-Q4_K_XL

List all available models

lemonade list

Hermes Agent

How to use unsloth/Qwen3.8-Flash-Next-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use unsloth/Qwen3.8-Flash-Next-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

New discussion

Resources

View closed (7)

Run 170 tokens/s Qwen3.8-Flash with MTP! ⚡

pinned

🔥❤️ 14

#56 opened 2 months ago by

danielhanchen

All Qwen3.8-Flash quants are here! 🔥 Run in Unsloth Desktop!

pinned

❤️ 18

#33 opened 2 months ago by

danielhanchen

Spelling issues on Q6_K_XL / Q5_K_XL

#89 opened 2 days ago by

auf1r2

Qwen3.8-Flash-Next-GGUF Unsloth Dynamic 3.0 Quants

👍 3

#88 opened 3 days ago by

Vinyl86

llama.cpp merge weights update

#87 opened 9 days ago by

TkskKurumi

UD-Q5_K_XL: shard 1 is 6 bytes short of its aligned data offset

#86 opened 15 days ago by

Vardaan97

BF16 part 3 out of 8 is damaged, its size 102gib.

#85 opened 22 days ago by

Issa2k5

mtp weight failed! check_tensor_dims: tensor 'token_embd.weight' not found

#84 opened 24 days ago by

Butterfly-314

MTP benchmarks

#83 opened 26 days ago by

rtzurtz

A quant in between Q4_K_XL and IQ4_XS

#82 opened 28 days ago by

Puchu1

Any way to separately select ngram quantization and model quantization?

#81 opened 28 days ago by

nawoalanor

unsloth studio vs strata engine for qwen3.8-flash-next

#80 opened about 1 month ago by

mayankiit04

llama.cpp now supports MTP in main, but these quants don't seem compatible

#79 opened about 1 month ago by

FlorinAndrei

will the new llama.cpp build 11330+ support the mtp file?

🔥👍 3

#78 opened about 1 month ago by

mayankiit04

ud-q5_k_s or ud-q5_k_m please

#77 opened about 1 month ago by

Mandu-Recs

Does this quant actually support engram SSD offload or not in llama.cpp??

#76 opened about 1 month ago by

EnderOLED

n-gram from SSD: how exactly?

#75 opened about 1 month ago by

dilavni

Running IQ4_XS with MTP , crashing after a while.

➕ 1

#74 opened about 1 month ago by

Oxidez

What is the actual difference between these two drafts? (MTP Q4_K_M | 2.79 GB) vs ( MTP Q4_K_M | 1.91 GB)

❤️ 1

#73 opened about 1 month ago by

sarkaritamminen

[BUG / Regression] SIGABRT in build_hc_mix during MTP graph reservation for Qwen3.8-Flash-Next (qwen4exp)

#72 opened about 2 months ago by

senigiro

llama cpp prebuilt 10909 and higher throw core exception

👍 1

#71 opened about 2 months ago by

linhcnguyen

Are there any qualitative benchmarks between Q_8 vs. BF16?

#70 opened about 2 months ago by

BigDeeper

单块16GB GPU + 64GB 内存：UD-IQ3_XXS（Windows）上27.5–29 tok/s — 完整指南 + 内置配置

#69 opened about 2 months ago by

Yue9527

Up to 34 t/s on 2× Tesla V100 16 GB + dual Xeon E5-2680 v4 (128 GB DDR4): tuning results for llama.cpp and ik_llama.cpp

🔥 5

#68 opened about 2 months ago by

Goydman

Request: a UD quant between IQ4_XS (93.7 GB) and Q4_K_XL (111 GB) for 128 GB unified-memory machines (Strix Halo)

👍 3

#67 opened about 2 months ago by

bitlamas

UD-IQ1_M on 6GB VRAM (GTX 1660 Ti): numbers, ladder to 128k, hiccups + fixes

#66 opened about 2 months ago by

TitanMamo10

Run UD-Q4-K-XL on 16GB VRAM and 32 GB RAM at 15+ tok/s

👀 1

#65 opened about 2 months ago by

Apolog1ze-Dev

Why are PLE/ngram weights quantized at all?

#64 opened about 2 months ago by

yurah

The best model under 200B reaches near-10 exact batch throughput on one laptop CPU

#63 opened about 2 months ago by

shyringo

Qwen3.8-Flash-Next UD-Q2_K_XL: deterministic exact-literal truncation with contrasting long-context fidelity vs Q2_K

#62 opened about 2 months ago by

nkoriyama

Quick share: Got 30–40 t/s on the RTX 5060 Ti 16GB

#61 opened about 2 months ago by

QilinWan

Too slow in 2 x H100, got decode 40t/s

#60 opened 2 months ago by

tientitmu

Small gain with MTP

#59 opened 2 months ago by

OmarColocci

Strange URL lookup when using agentic tools with it

#58 opened 2 months ago by

vico44

Claude code cli errors out with llama.cpp

#57 opened 2 months ago by

heapmaster

Tested new MTP module with UD-Q4_K_XL on Pro 6000 Max-Q

👍❤️ 10

#55 opened 2 months ago by

KouKou1986

MTP loading errors with IQ3_XXS quantz

➕👍 18

#54 opened 2 months ago by

vico44

can we have a gguf varity where ngram layer 2 is at q8?

#53 opened 2 months ago by

mayankiit04

Qwen3.8-Flash-Next (qwen4exp) garbage output on Strix Halo (gfx1151) over ROCm 7.x - root cause and verified fix (build from PR #27941)

👀👍 8

#52 opened 2 months ago by

k-mktr

Is it possible to made a GGUF without N-gram?

#51 opened 2 months ago by

BVEsun

How can I improve the prefilling speed for this model?

#49 opened 2 months ago by

BipedalBit

Getting 10+ tok/s on a 3060 laptop with 48GB RAM

🔥 12

#48 opened 2 months ago by

2vibeornot2vibe

context type MTP requested but model doesn't contain MTP layers

👀 4

#47 opened 2 months ago by

Manhandle6292

Is it possible to run it with a combined memory setup: 16 GB VRAM + 64 GB RAM + SSD for offloading n-grams?

#46 opened 2 months ago by

testtestad

Qwen3.8-Flash-Next on Strix Halo: ~40 tok/s code at 120K+ context

👍 9

#45 opened 2 months ago by

Engardium

Can the 51B n-gram table be kept on NVMe/SSD?

#44 opened 2 months ago by

NamerPRO

I'm a beginner, how do I use GGUF files that contain multiple files?

#43 opened 2 months ago by

jevencheung

Q4_K_XL / llama-server / Hallucination after 150k of filled context

#42 opened 2 months ago by

manisab

Works on my machine

🤯 2

#41 opened 2 months ago by

emcourtney

Qwen3.8-Flash-Next: deep-context decode slowdown + top_k crash + MTP results (3x RTX 3090, full log)

#40 opened 2 months ago by

karantula