Instructions to use agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-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 agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-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 agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF
# Run inference directly in the terminal:
llama cli -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF
# Run inference directly in the terminal:
llama cli -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF
# Run inference directly in the terminal:
./llama-cli -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF
# Run inference directly in the terminal:
./build/bin/llama-cli -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

Use Docker

docker model run hf.co/agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

LM Studio
Jan

vLLM

How to use agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-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": "agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

Ollama
How to use agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF with Ollama:
```
ollama run hf.co/agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF
```
Unsloth Desktop

How to use agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

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": "agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF with Docker Model Runner:
```
docker model run hf.co/agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF
```

Lemonade

How to use agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

Run and chat with the model

lemonade run user.Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-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 agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

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 agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

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 "agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF" \
  --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 ROCmFP4-FAST imatrix GGUF

The only quant of this model that fits fully in VRAM on a 128 GB unified-memory device without giving up quality: 87.06 GiB, 4.23 bpw, within 2.5% perplexity of the unquantized model (see below) — and it beats every alternative under 110 GiB we could find. Sized for the 96 GiB VRAM carve-out of a Strix Halo (Ryzen AI MAX+ 395 / Radeon 8060S). The whole model runs resident on GPU: nothing falls back to host RAM or CPU compute, not the experts, not the 51.2 B-parameter n-gram table.

Two layouts, same weights, same size (87.06 GiB either way — splitting the n-gram table per head is a byte-for-byte restructuring, not a re-quantize):

root — n-gram table split per head, fully VRAM-resident.
joined/ — n-gram table as one tensor, portable; needs --ngram-on-disk or host RAM for that tensor, since a single tensor that size is past what most Vulkan devices accept as one buffer.

Setup

Qwen3.8-Flash-Next itself is upstream (ggml-org/llama.cpp#27742). This fork is still needed for the ROCmFPx quant types and the per-head PLE layout.

git clone https://github.com/LaurentZuijdwijk/llama.cpp
cd llama.cpp && git checkout vulkan/qwen4exp-rocmfpx
cmake -B build -DGGML_VULKAN=ON -DCMAKE_BUILD_TYPE=Release
cmake --build build -j

Run

# per-head table, fully on the GPU
./build/bin/llama-cli \
  -m Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix.gguf \
  -ngl 99 -c 32768

# joined table, off the GPU and off host RAM
./build/bin/llama-cli \
  -m joined/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-joined.gguf \
  -ngl 99 --ngram-on-disk --ngram-cache 8192 -c 32768

--ngram-cache defaults to 256 MiB; raise it for long generations or throughput drops off over the course of a conversation.

The two layouts are interchangeable: gguf_split_ple_heads.py converts one to the other by copying quantized bytes verbatim — no dequantize, no requantize, no quality change.

Perplexity

wikitext-2 raw, 145 chunks at -c 2048.

build	PPL	vs. reference
unquantized reference (as reported in PR 27742)	4.0068 +/- 0.02271	-
this file	4.1062 +/- 0.02329	+2.48%

Against AesSedai's quants, compared the fair way (each build's PPL against its own measured reference, since their test methodology differs from ours):

build	size	PPL ratio vs. own reference
AesSedai IQ3_S	107.38 GiB	+6.10%
this file	87.06 GiB	+2.48%
AesSedai IQ4_XS	117.13 GiB	+3.12%
AesSedai Q4_K_M	135.38 GiB	+0.61%

Beats their IQ4_XS and IQ3_S on quality at a smaller size.

MTP draft head

Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST-GGUF holds the model's own multi-token-prediction head, exported from the same checkpoint. Currently mismatched with this file's expert precision (draft is flat FP4, this file's gate/up are Q4_K) — needs requantizing before it's worth enabling here.

Credits

qwen4exp support is the work of Daniel Han (@danielhanchen), from ggml-org/llama.cpp#27742, merged upstream. This fork is only still needed for what's listed under Setup above.

Quant formats hand-ported from ciru-ai/ROCmFPX. Calibration corpora from bartowski and Thireus, credited above. Base model by the Qwen team.

Not included

Vision tower.

License

Qwen Community License 1.0, included as LICENSE.

Downloads last month: 421

GGUF

Model size

177B params

Architecture

qwen4exp

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Model tree for agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF

Base model

Qwen/Qwen3.8-Flash-Next

Quantized

(126)

this model