Instructions to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF",
	filename="Qwen3.5-9B-NVFP4-MTP-GGUF.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4
# Run inference directly in the terminal:
llama-cli -hf michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4
# Run inference directly in the terminal:
llama-cli -hf michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

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 michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4
# Run inference directly in the terminal:
./llama-cli -hf michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

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 michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4
# Run inference directly in the terminal:
./build/bin/llama-cli -hf michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

Use Docker

docker model run hf.co/michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

LM Studio
Jan
Ollama
How to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with Ollama:
```
ollama run hf.co/michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4
```

Unsloth Studio

How to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF to start chatting

How to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

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 michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

Run Hermes

hermes

Atomic Chat new
Docker Model Runner
How to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with Docker Model Runner:
```
docker model run hf.co/michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4
```

Lemonade

How to use michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull michaelw9999/Qwen3.5-9B-NVFP4-MTP-GGUF:NVFP4

Run and chat with the model

lemonade run user.Qwen3.5-9B-NVFP4-MTP-GGUF-NVFP4

List all available models

lemonade list

This Qwen3.5-9B model was quantized with NVFP4 with MTP support, using my soon to be released NVFP4 GGUF quantizer. This autotunes the model to reduce ppl and kld as much as possible, and selects an optimal llama.cpp tensor distribution.

This keeps the model size down to just 5.66GB (previously 6.21GB) while still improving quality, and MTP increases speed.

Updated results with better ppl/kld distribution:

====== Perplexity statistics ======
Mean PPL(Q)                   :   8.252679 ±   0.055494
Mean PPL(base)                :   8.184061 ±   0.055382
Cor(ln(PPL(Q)), ln(PPL(base))):  98.68%
Mean ln(PPL(Q)/PPL(base))     :   0.008349 ±   0.001098
Mean PPL(Q)/PPL(base)         :   1.008384 ±   0.001108
Mean PPL(Q)-PPL(base)         :   0.068618 ±   0.009021

====== KL divergence statistics ======
Mean    KLD:   0.061969 ±   0.000836
Maximum KLD:  25.706320
99.9%   KLD:   3.060981
99.0%   KLD:   0.504558
95.0%   KLD:   0.178821
90.0%   KLD:   0.113534
Median  KLD:   0.027535
10.0%   KLD:   0.001330
 5.0%   KLD:   0.000385
 1.0%   KLD:   0.000052
 0.1%   KLD:   0.000007
Minimum KLD:  -0.000136

====== Token probability statistics ======
Mean    Δp: -0.496 ± 0.017 %
Maximum Δp: 99.953%
99.9%   Δp: 49.024%
99.0%   Δp: 16.581%
95.0%   Δp:  7.308%
90.0%   Δp:  4.173%
75.0%   Δp:  0.814%
Median  Δp: -0.013%
25.0%   Δp: -1.507%
10.0%   Δp: -5.672%
 5.0%   Δp: -9.537%
 1.0%   Δp: -21.667%
 0.1%   Δp: -51.141%
Minimum Δp: -99.820%
RMS Δp    :  6.656 ± 0.056 %
Same top p: 89.216 ± 0.081 %
Top flip weight:   0.009668
Top prob RMSE  :   0.080021
Entropy RMSE   :   0.272950

Previous results of this model (older quantizer):

====== Perplexity statistics ======
Mean PPL(Q)                   :   8.511450 ±   0.057807
Mean PPL(base)                :   8.184061 ±   0.055382
Cor(ln(PPL(Q)), ln(PPL(base))):  98.30%
Mean ln(PPL(Q)/PPL(base))     :   0.039224 ±   0.001251
Mean PPL(Q)/PPL(base)         :   1.040003 ±   0.001301
Mean PPL(Q)-PPL(base)         :   0.327389 ±   0.010714

====== KL divergence statistics ======
Mean    KLD:   0.082267 ±   0.000935
Maximum KLD:  26.195276
99.9%   KLD:   3.699525
99.0%   KLD:   0.655555
95.0%   KLD:   0.239608
90.0%   KLD:   0.155213
Median  KLD:   0.038994
10.0%   KLD:   0.001889
 5.0%   KLD:   0.000544
 1.0%   KLD:   0.000079
 0.1%   KLD:   0.000011
Minimum KLD:  -0.000016

====== Token probability statistics ======
Mean    Δp: -0.911 ± 0.020 %
Maximum Δp: 99.991%
99.9%   Δp: 50.329%
99.0%   Δp: 17.867%
95.0%   Δp:  8.117%
90.0%   Δp:  4.473%
75.0%   Δp:  0.746%
Median  Δp: -0.044%
25.0%   Δp: -2.061%
10.0%   Δp: -7.160%
 5.0%   Δp: -11.774%
 1.0%   Δp: -27.189%
 0.1%   Δp: -62.353%
Minimum Δp: -99.881%
RMS Δp    :  7.690 ± 0.057 %
Same top p: 87.429 ± 0.086 %

Compare the original Qwen3.5-9B-NVFP4(made via ModelOpt):

====== Perplexity statistics ======
Mean PPL(Q)                   :   8.676226 ±   0.059423
Mean PPL(base)                :   8.184061 ±   0.055382
Cor(ln(PPL(Q)), ln(PPL(base))):  98.30%
Mean ln(PPL(Q)/PPL(base))     :   0.058398 ±   0.001259
Mean PPL(Q)/PPL(base)         :   1.060137 ±   0.001334
Mean PPL(Q)-PPL(base)         :   0.492165 ±   0.011330

====== KL divergence statistics ======
Mean    KLD:   0.085222 ±   0.000895
Maximum KLD:  27.925140
99.9%   KLD:   3.518223
99.0%   KLD:   0.672682
95.0%   KLD:   0.250890
90.0%   KLD:   0.164238
Median  KLD:   0.042011
10.0%   KLD:   0.001942
 5.0%   KLD:   0.000554
 1.0%   KLD:   0.000077
 0.1%   KLD:   0.000011
Minimum KLD:  -0.000043

====== Token probability statistics ======
Mean    Δp: -0.996 ± 0.020 %
Maximum Δp: 99.998%
99.9%   Δp: 48.766%
99.0%   Δp: 18.332%
95.0%   Δp:  8.174%
90.0%   Δp:  4.538%
75.0%   Δp:  0.727%
Median  Δp: -0.053%
25.0%   Δp: -2.165%
10.0%   Δp: -7.505%
 5.0%   Δp: -12.305%
 1.0%   Δp: -27.862%
 0.1%   Δp: -63.063%
Minimum Δp: -99.966%
RMS Δp    :  7.856 ± 0.056 %
Same top p: 87.312 ± 0.087 %