Instructions to use localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF",
	filename="Qwen3.5-4B-MTP-Q4_K_M.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 localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_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 localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_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 localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M

Use Docker

docker model run hf.co/localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF with Ollama:
```
ollama run hf.co/localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M
```

Unsloth Studio

How to use localweights/Qwen3.5-4B-MTP-Q4_K_M-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 localweights/Qwen3.5-4B-MTP-Q4_K_M-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 localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF to start chatting

How to use localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M

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": "localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use localweights/Qwen3.5-4B-MTP-Q4_K_M-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 localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_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 localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat new
Docker Model Runner
How to use localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF with Docker Model Runner:
```
docker model run hf.co/localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M
```

Lemonade

How to use localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3.5-4B-MTP-Q4_K_M-GGUF-Q4_K_M

List all available models

lemonade list

Qwen3.5-4B-MTP-Q4_K_M-GGUF

Qwen3.5-4B (qwen35 dense+ssm hybrid arch) with NextN/MTP head preserved, quantized to Q4_K_M (5.25 BPW). Built via the patched convert_hf_to_gguf.py from patched llama.cpp build with Qwen3.5/3.6 MTP support.

Sibling: localweights/Qwen3.5-4B-MTP-IQ4_XS-GGUF (smaller, faster, 4.25 BPW).

Files

File	Size	Purpose
`Qwen3.5-4B-MTP-Q4_K_M.gguf`	2.7 GB	Q4_K_M with 3 outlier tensors kept at `bf16`.

Build pipeline

python convert_hf_to_gguf.py /path/to/Qwen3.5-4B \
    --outfile Qwen3.5-4B-MTP-bf16.gguf

# 3 tensors have absurdly large values (1e19 to 1e36) that overflow Q4_K's
# fp16 scale-block storage. Keep them at bf16 to pass row-data validation.
llama-quantize \
  --tensor-type "blk\.9\.ssm_out\.weight=bf16" \
  --tensor-type "blk\.15\.attn_output\.weight=bf16" \
  --tensor-type "blk\.24\.ffn_gate\.weight=bf16" \
  Qwen3.5-4B-MTP-bf16.gguf \
  Qwen3.5-4B-MTP-Q4_K_M.gguf \
  Q4_K_M

Optimal serving config

llama-server -m Qwen3.5-4B-MTP-Q4_K_M.gguf \
  -ngl 999 -fa on \
  --spec-type mtp --spec-draft-n-max 2 \
  --no-mmap \
  --ctx-size 8192 -ctk q4_0 -ctv q4_0 \
  --parallel 1 --kv-unified \
  --metrics --jinja

Performance — `--spec-draft-n-max` sweep

Measured 2026-05-06 on Crucible (9950X, 96 GB DDR5-4800 dual-channel, RTX 3090 Ti). Prompt: count 1→50, 300-token decode.

GPU (RTX 3090 Ti, full offload)

n	Decode tok/s	Accept rate
2	271 ← peak	100%
3	273	90%

CPU (16 threads, 2-channel DDR5-4800)

n	Decode tok/s	Accept rate
1	24.2	100%
2	30.4 ← peak	100%
3	31.1	87%
4	30.9	75%
5	25.2	55%

Vs sibling IQ4_XS at peak: 271 t/s GPU (vs 289), 30.4 t/s CPU (vs 35.9). Q4_K_M trades ~6–15% throughput for ~24% more bits per weight.

Tokenizer

qwen35 pre-tokenizer, 151,936 vocab. Standard chat template.

License

Apache 2.0.

Provenance

Built on Crucible: 9950X / 96 GB DDR5 / RTX 3090 Ti. Sibling repos: localweights/Qwen3.5-4B-MTP-IQ4_XS-GGUF, localweights/Qwen3.6-{27B,35B-A3B}-MTP-IQ4_XS-GGUF.

Downloads last month: 793

GGUF

Model size

4B params

Architecture

qwen35

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for localweights/Qwen3.5-4B-MTP-Q4_K_M-GGUF

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Quantized

(235)

this model