Instructions to use h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-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 h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-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 h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL
# Run inference directly in the terminal:
llama cli -hf h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL
# Run inference directly in the terminal:
llama cli -hf h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

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 h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL
# Run inference directly in the terminal:
./llama-cli -hf h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

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 h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

Use Docker

docker model run hf.co/h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

LM Studio
Jan
Ollama
How to use h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF with Ollama:
```
ollama run hf.co/h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL
```

Unsloth Studio

How to use h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-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 h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-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 h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF to start chatting

How to use h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

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": "h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-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 h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

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 h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

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 "h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL" \
  --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"

Docker Model Runner
How to use h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF with Docker Model Runner:
```
docker model run hf.co/h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL
```

Lemonade

How to use h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull h4rm0n1c/qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF:IQ4_NL

Run and chat with the model

lemonade run user.qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF-IQ4_NL

List all available models

lemonade list

qwen3.5-24b-a3b-reap-0.32-iq4_nl-GGUF / README.md

h4rm0n1c

Update README.md

2a177e8 verified about 2 months ago

preview code

Raw

History Blame Contribute Delete

2.89 kB

	---
	license: apache-2.0
	library_name: gguf
	base_model: sandeshrajx/Qwen3.5-24B-A3B-REAP-0.32
	tags:
	- qwen3.5
	- moe
	- gguf
	- iq4_nl
	- reap
	- pruned
	- quantzhai
	- quantized
	---

	# Qwen3.5-24B-A3B-REAP-0.32 IQ4_NL GGUF

	Quanter's Note: This model is the parent of every solid performing reasoning distill I've benched out of 40 or so models in the last month, this thing has some solid potential for a model that has been given a partial lobotomy, awfully impressive.
	Perplexity test I devised is Extremely Tough on models, high convergence rate on alien/unusual codebases, plenty of potential for finetunes.
	I felt the base model deserved some attention alongside its author and I'm glad to see people downloading it.
	Every finetune of this model I benched, IMPROVED over this base model. Very impressive.

	GGUF quantization of [sandeshrajx/Qwen3.5-24B-A3B-REAP-0.32](https://huggingface.co/sandeshrajx/Qwen3.5-24B-A3B-REAP-0.32) — a REAP-pruned 24B total, ~3B active MoE model.

	REAP (Razor Edge And Pruning, [arxiv:2510.13999](https://arxiv.org/abs/2510.13999)) is a structured pruning technique that reduces the base Qwen3.5 model while preserving capability.

	Source: [sandeshrajx/Qwen3.5-24B-A3B-REAP-0.32](https://huggingface.co/sandeshrajx/Qwen3.5-24B-A3B-REAP-0.32)
	Converted by: [QuantZhai](https://github.com/h4rm0n1c/quantzhai) benchmark pipeline
	Quantization: IQ4_NL with importance matrix (imatrix from sandeshrajx)

	## Model Details

	\| Property \| Value \|
	\|---\|---\|
	\| Architecture \| Qwen3.5 MoE, REAP-pruned \|
	\| Parameters \| 24B total, ~3B active \|
	\| Blocks \| 40 \|
	\| Experts \| 175 (REAP-split), 8 active per token \|
	\| Context length \| 262144 (256K) \|
	\| Hidden size \| 3072 \|
	\| Attention heads \| 32, KV heads = 2 \|
	\| Quantization \| IQ4_NL (4.58 bpw) with imatrix \|
	\| File size \| 14.0 GB \|

	## Benchmarks

	Hardware: dual-GPU (RTX 3080 10GB + V100-SXM2 32GB)
	Engine: llama.cpp with TurboQuant KV (q8_0 K / turbo3 V)
	Perplexity: [`macvox68`](https://github.com/h4rm0n1c/macvox68) code corpus, ctx=4096, stride=512

	\| Metric \| Cold \| Warm \|
	\|---\|---\|---\|
	\| PPL \| 3.0205 \| 1.8298 \|
	\| TPS \| 33.6 tok/s \| 45.9 tok/s \|
	\| TTFT \| 1486 ms \| 1090 ms \|

	### QuantZhai Ranking

	Rank #15 of 47 — combined score 63.6 (equal-weight: TPS, PPL, convergence).
	Higher than many larger dense models — REAP pruning + IQ4_NL quantization is an efficient combination.

	## Usage

	```bash
	llama-cli -m Qwen3.5-24B-A3B-REAP-0.32-IQ4_NL.gguf \
	-p "Write a mergesort in Python" \
	-n 1024 -t 12 --temp 0.6 --top-p 0.95

	llama-server -m Qwen3.5-24B-A3B-REAP-0.32-IQ4_NL.gguf \
	--host 0.0.0.0 --port 8080 -ngl 99 -t 12 \
	--cache-type-k q8_0 --cache-type-v turbo3
	```

	## License

	Apache 2.0 (this quantization).
	Source model by [sandeshrajx](https://huggingface.co/sandeshrajx) under Apache 2.0 — see [arxiv:2510.13999](https://arxiv.org/abs/2510.13999).