Instructions to use matchachoco/Qwen3.5-4B-QuantStudy-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 matchachoco/Qwen3.5-4B-QuantStudy-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 matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS
# Run inference directly in the terminal:
llama cli -hf matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS
# Run inference directly in the terminal:
llama cli -hf matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

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 matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS
# Run inference directly in the terminal:
./llama-cli -hf matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

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 matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS
# Run inference directly in the terminal:
./build/bin/llama-cli -hf matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

Use Docker

docker model run hf.co/matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

LM Studio
Jan

vLLM

How to use matchachoco/Qwen3.5-4B-QuantStudy-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "matchachoco/Qwen3.5-4B-QuantStudy-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": "matchachoco/Qwen3.5-4B-QuantStudy-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

Ollama
How to use matchachoco/Qwen3.5-4B-QuantStudy-GGUF with Ollama:
```
ollama run hf.co/matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS
```
Unsloth Desktop

How to use matchachoco/Qwen3.5-4B-QuantStudy-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

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": "matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use matchachoco/Qwen3.5-4B-QuantStudy-GGUF with Docker Model Runner:
```
docker model run hf.co/matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS
```

Lemonade

How to use matchachoco/Qwen3.5-4B-QuantStudy-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

Run and chat with the model

lemonade run user.Qwen3.5-4B-QuantStudy-GGUF-IQ4_XS

List all available models

lemonade list

Hermes Agent

How to use matchachoco/Qwen3.5-4B-QuantStudy-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 matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

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 matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use matchachoco/Qwen3.5-4B-QuantStudy-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS

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 "matchachoco/Qwen3.5-4B-QuantStudy-GGUF:IQ4_XS" \
  --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"

Qwen3.5-4B-QuantStudy-GGUF

수식과 표로 읽는 실험 보고서 / Read the report · Code and evaluation evidence

Qwen3.5-4B 전체 모델 양자화 비교에서 실제로 저장·재로딩·평가한 결과물입니다. These are the exact packed artifacts evaluated in the 2026-09-22 quantization study, not newly retuned checkpoints.

Scope and format

Base: Qwen/Qwen3.5-4B, revision 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a.
Common targets: 200 text Linear matrices, 3,565,158,400 original weights (96 FFN, 32 full-attention, 72 gated-delta projections).
Embedding/tied output head, norms, biases, convolution/state tensors and the 48 small gated-delta in_proj_a/b matrices retain high precision. AWQ can rescale preserved tensors while retaining high precision.
모든 가중치가 4비트인 모델이 아닙니다. This custom preservation policy differs from common default community recipes bearing similar names.

Text-only GGUF; no vision projector is distributed. Tested with llama.cpp b10690 (bdf3955159d7184f44b76091973eeff532890a35).

Files and recipe

Candidate	Settings	Weight file	Decimal GB
K0	Q4_K, pure, no importance matrix	Qwen3.5-4B-K0-Q4_K.gguf	3.299
K1	Q4_K, pure, calibration importance matrix	Qwen3.5-4B-K1-Q4_K.gguf	3.299
KM	Q4_K_M + importance matrix; 168 Q4_K / 32 Q6_K target matrices	Qwen3.5-4B-KM-Q4_K_M.gguf	3.464
I4	IQ4_XS, pure, calibration importance matrix	Qwen3.5-4B-I4-IQ4_XS.gguf	3.188

Q4_K uses 32 weights per subblock and 256 per superblock. IQ4_XS uses its own nonuniform encoding. K0 still uses llama.cpp fitting; it is not the RTN candidate.

Measured quality

같은 엔진의 BF16 기준과 비교합니다. NLL은 정답 토큰에 부여한 확률의 음의 로그 평균으로, 낮을수록 좋으며 정답률과는 다릅니다.

Model	NLL ↓	PPL ↓	Δ NLL vs own BF16 ↓	KMMLU %	ARC-C norm %	IFEval strict %
B0 (native)	1.851498	6.3694	+0.000000	48.2	56.2	82.62
K0 (native)	1.886050	6.5933	+0.034553	49.4	53.0	80.41
K1 (native)	1.874492	6.5175	+0.022994	46.6	54.4	80.96
KM (native)	1.869853	6.4873	+0.018355	48.8	53.8	79.85
I4 (native)	1.881741	6.5649	+0.030243	47.8	55.4	81.89

Calibration: 128 windows × 1,024 tokens (64 Korean / 32 English / 32 code source documents); validation 32 and final 128 documents are disjoint by exact document SHA.
Final NLL: 130,944 scored tokens. KMMLU: fixed stratified 500-item subset, not the official whole benchmark. ARC-C: fixed 500 items, character-length-normalized choice scores. IFEval: 541 prompts, prompt strict, greedy, thinking off, maximum 2,048 output tokens.
Final data did not select recipe settings. Failures were not removed from denominators. Semantic duplication and pretraining exposure are not ruled out.
One calibration seed/configuration; not a claim that the algorithms were exhaustively optimized. See paired confidence intervals and failure cases before interpreting small differences.

Download and run

hf download matchachoco/Qwen3.5-4B-QuantStudy-GGUF --local-dir ./Qwen3.5-4B-QuantStudy-GGUF

For an exact immutable version, add --revision <full commit SHA> shown in the repository history. File hashes are in SHA256SUMS and ARTIFACT-MANIFEST.json.

Example with the evaluated llama.cpp release:

llama-cli -m ./Qwen3.5-4B-QuantStudy-GGUF/Qwen3.5-4B-KM-Q4_K_M.gguf -ngl 99 -c 8192 -cnv

Choose another listed GGUF filename to compare another candidate. This is a convenience chat example; the study uses the fixed prompts in the report. Long-context and vision quality were not evaluated.

Reproducibility and attribution

See fixed environment, data and target policy, quantization-study.json, and the full report. Weight bytes and saved quantization configs are preserved from the completed experiment. Publishing adds documentation and checksums; it does not change model weights.

Base model by the Qwen team. Derived weights are distributed with the upstream Apache-2.0 LICENSE; see NOTICE for modifications. This is a learning/portfolio experiment, not an official Qwen release or a newly proposed quantization algorithm. Coding, execution automation, evaluation aggregation and documentation used Codex assistance.

Downloads last month: 288

GGUF

Model size

4B params

Architecture

qwen35

Hardware compatibility

4-bit

Model tree for matchachoco/Qwen3.5-4B-QuantStudy-GGUF

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Quantized

(443)

this model