Instructions to use tawei-amd/Qwen3.8-Flash-Next-strixflash-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 tawei-amd/Qwen3.8-Flash-Next-strixflash-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 tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0
# Run inference directly in the terminal:
llama cli -hf tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0
# Run inference directly in the terminal:
llama cli -hf tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

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 tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0
# Run inference directly in the terminal:
./build/bin/llama cli -hf tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

Use Docker

docker model run hf.co/tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

Llama new
LM Studio
Jan

vLLM

How to use tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tawei-amd/Qwen3.8-Flash-Next-strixflash-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": "tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

Ollama
How to use tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF with Ollama:
```
ollama run hf.co/tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0
```
Unsloth Desktop

How to use tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

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": "tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF with Docker Model Runner:
```
docker model run hf.co/tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0
```

Lemonade

How to use tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

Run and chat with the model

lemonade run user.Qwen3.8-Flash-Next-strixflash-GGUF-Q8_0

List all available models

lemonade list

Hermes Agent

How to use tawei-amd/Qwen3.8-Flash-Next-strixflash-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 tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

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 tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0

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 "tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF:Q8_0" \
  --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"

Qwen3.8-Flash-Next, strix-flash mix A (GGUF)

Quantized weights of Qwen/Qwen3.8-Flash-Next for local serving on AMD Strix Halo (Ryzen AI Max, gfx1151) with Halocpp. They are made for, and tested with, the strix-flash engine that Halocpp runs for this model.

./halocpp pull qwen3.8-flash-next
./halocpp serve qwen3.8-flash-next          # OpenAI-compatible API on 127.0.0.1:52625

Files

file	bytes	sha256
`LICENSE`	3,235	`a0dc422560841fd68e06d974907f8b4c709bca44a67daad2b528437bdf676c08`
`mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf`	2,786,568,256	`5ff54097406a905cf3a724c709124ceb0e3e10235ee862298969e91c96fa96e6`
`Qwen3.8-Flash-Next-strixflash-mixA-00001-of-00006.gguf`	375,531,936	`ac483373b2c194c2ed73e3d428a6c577dc294fc54e877972db033697ce10aa3e`
`Qwen3.8-Flash-Next-strixflash-mixA-00002-of-00006.gguf`	28,800,138,432	`0e9027aa7de4817c91b5954039b563484466392e29a6b051b096f9c47a71a1c6`
`Qwen3.8-Flash-Next-strixflash-mixA-00003-of-00006.gguf`	19,790,357,632	`5e1d95f7a3e6a60e9b7da64beea5b289228edc4cd97341509e93592b8ee9a65b`
`Qwen3.8-Flash-Next-strixflash-mixA-00004-of-00006.gguf`	19,608,645,248	`64f0284324de3f8e0fa15856cfc575b9d5a261ce399a6c9d2fd3ca596a36acf6`
`Qwen3.8-Flash-Next-strixflash-mixA-00005-of-00006.gguf`	19,719,695,840	`93a0d6ee5205363965d9965c8a5c072241fe6a5141fc9fff12fab74690531fc5`
`Qwen3.8-Flash-Next-strixflash-mixA-00006-of-00006.gguf`	4,103,658,848	`ce1ce2d8bb8e73f761694768d5dc5edef1270a8bb960dd6515f9aa1086b24c9d`
`tokenizer/chat_template.jinja`	8,952	`c3cf9e34abf4f9e36c2d72165aa9c132d3e2a725b6c2586aaa3a8af9d7a81041`
`tokenizer/config.json`	4,745	`889658f2508e8c61d409b02e70e0d78d8d4452ec65aaafbe129805d213d2e74b`
`tokenizer/generation_config.json`	202	`e70c136c1b78ddc1fb0905bac8e733a4dc448d4f852a5dd75143fffc70be550e`
`tokenizer/tokenizer_config.json`	17,928	`b11349aafa7cdc6a320767cf7ceb29ed82f7eda5d65e8e0819e76f0ce947bf27`
`tokenizer/tokenizer.json`	12,809,320	`0997f410c57a1f4e53b09e4be8f4a172d90edd9564368fb0847030937229b9f3`

Qwen3.8-Flash-Next-strixflash-mixA-*.gguf is one model split into GGUF shards with llama.cpp's llama-gguf-split (load the first shard; the others are found by name). SHA256SUMS lists every file.

How mix A was made

Starting from unsloth's UD-IQ4_XS GGUF of the model (unsloth/Qwen3.8-Flash-Next-GGUF, revision 38bb39ee97821de2c9009abb7e93950eec396e66), the dense trunk was re-quantized with GPTQ; the routed experts are kept as published:

tensors	format
DeltaNet input projections, attention q / k / v	Q5_0, GPTQ
DeltaNet output, shared experts, LM head	IQ4_NL, GPTQ
hyper-connections, attention output	Q8_0
routed experts	as published: IQ3_S gate / up, IQ4_NL down (Q8_0 in 5 layers)

GPTQ used Hessians collected by the engine from 48 calibration sequences, alternating wikitext-2 test text beyond its first 100,000 tokens and C/C++ source code, so the calibration text is disjoint from the perplexity evaluation below. About 4.4 GB of weights are read per decoded token.

mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf is the model's multi-token-prediction block, used as the draft head for speculative decoding; it is unsloth's file, unchanged. The tokenizer/ files are the base model's, unchanged (revision de4b8e4d43b917e7706784d8bb445c9af86a3540).

Accuracy

Perplexity on 32,704 wikitext-2 test tokens (64 sequences of 512, decode path): measurement pending.

License

These weights are a derivative of Qwen3.8-Flash-Next and are distributed under the Qwen Community License 1.0, included as LICENSE. Its conditions apply to every use of these files.

Downloads last month: 1,265

GGUF

Model size

177B params

Architecture

qwen4exp

Hardware compatibility

8-bit

View +1 variant

Model tree for tawei-amd/Qwen3.8-Flash-Next-strixflash-GGUF

Base model

Qwen/Qwen3.8-Flash-Next

Quantized

(530)

this model