Instructions to use ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-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 ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-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 ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M
# Run inference directly in the terminal:
llama cli -hf ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M
# Run inference directly in the terminal:
llama cli -hf ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

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 ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M
# Run inference directly in the terminal:
./llama-cli -hf ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

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 ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

Use Docker

docker model run hf.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

Llama new
LM Studio
Jan

vLLM

How to use ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-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": "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-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/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

Ollama
How to use ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF with Ollama:
```
ollama run hf.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M
```
Unsloth Desktop

How to use ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

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": "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF with Docker Model Runner:
```
docker model run hf.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M
```

Lemonade

How to use ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

Run and chat with the model

lemonade run user.Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF-IQ1_M

List all available models

lemonade list

Hermes Agent

How to use ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-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 ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

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 ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M

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 "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF:IQ1_M" \
  --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"

MTP / Expert Selection Map

#21

by Davidmg0815 - opened 26 days ago

Discussion

Davidmg0815

26 days ago

Hallo zusammen!

Der MTP-Head lässt sich nicht mit der geprunten Coder-Version laden, da der Head 512 Experten erwartet, das Modell aber nur 256 hat:
llama_model_load: error: expert_count mismatch (model: 256, mtp: 512)

Könntet ihr bitte entweder:

Einen passenden 256-Expert MTP-Head hochladen?
Oder die mask.pt (Expert Selection Map) bereitstellen, damit wir den Head selbst anpassen können?

Vielen Dank für eure tolle Arbeit an den Modellen!

sebastavar

25 days ago

+1 on this request. Some numbers from an M5 Max Mac Studio, running llama-server with n-gram drafting (ngram-map-k):

Writing new text: 47 to 54 tok/s
Editing existing code: 157 to 180 tok/s, with 93% of drafts accepted
Qwen3.8-27B at 8-bit with MTP on the same machine: 45 to 52 tok/s

So the Coder model already matches or beats the dense 27B with MTP, using only n-gram drafting. A native MTP head would help most on new text, where n-gram drafting has little to copy from.

One note on option 2: as far as I can tell, the MTP layer has its own expert set, separate from the 48 pruned layers, so the existing expert mask may not cover it. A 256-expert MTP head selected with the same RCO search (option 1) seems like the more reliable path.

Thanks for the release. It's now in my daily coding setup on a trial basis, and I'm happy to test an MTP build.

eudaimonent

10 days ago

Try this:

https://huggingface.co/eudaimonent/mtp-Qwen3.8-Flash-Next-Q4_0-256router

Davidmg0815

10 days ago

https://huggingface.co/Davidmg0815/Qwen3.8-Flash-Next-MTP-256exp-GGUF

Davidmg0815

10 days ago

Ich hab's jetzt so gelöst und damit die best n Ergebnisse... Ich hab über 100 token mit Ner alten 3080 Viel spaass

eudaimonent

9 days ago

What an excellent read on your model! I'm flattered that you tried out my mtp gguf. I haven't had the chance to dig up the rest of it. At the time, I didn't see anything that seemed to fit the bill, so I had the ISTA model whip a few things up for me. That resulted in this:

https://github.com/eudaimonenterprises/llama.cpp/tree/fix/mtp-hybrid-device-matching-exact

I'm working with a smaller dgpu and an igpu that I can get to work together over rpc. As you stated, memory is worth considering in that situation, hence the want to free up as much of it as I could. I'm definitely going to see how the other models work on my rig as I move forward. You highlighted some settings that I wasn't even aware of as I'm more of a generalist when it comes to these things. What I lack in depth I try to make up for in breadth. Such a great learning experience!

Davidmg0815

9 days ago

Thanks a lot — glad the write-up was useful! Your MTP GGUF was one of the more interesting ones to test.

Your branch fix/mtp-hybrid-device-matching-exact caught my eye: we run the Flash-Next 176B (hybrid, split across two RTX 3080) and MTP there gets unstable at large context. What exactly does your fix change — the device placement of the MTP head relative to the hybrid layers? If it's in a reviewable state I'd be happy to try it on our setup and report back.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment