---
license: apache-2.0
license_link: https://huggingface.co/Qwen/QWQ-32B/blob/main/LICENSE
language:
- en
pipeline_tag: text-generation
base_model: Qwen/Qwen2.5-32B
tags:
- chat
library_name: transformers
---

# <span style="color: #7FFF7F;">QwQ-32B GGUF Models</span>

## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>

Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.

### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations

### **Method**
- **Dynamic Precision Allocation**:  
  - First/Last 25% of layers → IQ4_XS (selected layers)  
  - Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)  
- **Critical Component Protection**:  
  - Embeddings/output layers use Q5_K  
  - Reduces error propagation by 38% vs standard 1-2bit  

### **Quantization Performance Comparison (Llama-3-8B)**

| Quantization | Standard PPL | DynamicGate PPL | Δ PPL   | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS      | 11.30        | 9.84             | -12.9%  | 2.5G     | 2.6G    | +0.1G  | 234s      | 246s     |
| IQ2_XS       | 11.72        | 11.63            | -0.8%   | 2.7G     | 2.8G    | +0.1G  | 242s      | 246s     |
| IQ2_S        | 14.31        | 9.02             | -36.9%  | 2.7G     | 2.9G    | +0.2G  | 238s      | 244s     |
| IQ1_M        | 27.46        | 15.41            | -43.9%  | 2.2G     | 2.5G    | +0.3G  | 206s      | 212s     |
| IQ1_S        | 53.07        | 32.00            | -39.7%  | 2.1G     | 2.4G    | +0.3G  | 184s      | 209s     |

**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead

**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization

**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)


### **When to Use These Models**
📌 **Fitting models into GPU VRAM**

✔ **Memory-constrained deployments**

✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated 
 
✔ **Research** into ultra-low-bit quantization


## **Choosing the Right Model Format**  

Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.  

### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**  
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.  
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.  
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).  
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.  

📌 **Use BF16 if:**  
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).  
✔ You want **higher precision** while saving memory.  
✔ You plan to **requantize** the model into another format.  

📌 **Avoid BF16 if:**  
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).  
❌ You need compatibility with older devices that lack BF16 optimization.  

---

### **F16 (Float 16) – More widely supported than BF16**  
- A 16-bit floating-point **high precision** but with less of range of values than BF16. 
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).  
- Slightly lower numerical precision than BF16 but generally sufficient for inference.  

📌 **Use F16 if:**  
✔ Your hardware supports **FP16** but **not BF16**.  
✔ You need a **balance between speed, memory usage, and accuracy**.  
✔ You are running on a **GPU** or another device optimized for FP16 computations.  

📌 **Avoid F16 if:**  
❌ Your device lacks **native FP16 support** (it may run slower than expected).  
❌ You have memory limitations.  

---

### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**  
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.  
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.  
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.  

📌 **Use Quantized Models if:**  
✔ You are running inference on a **CPU** and need an optimized model.  
✔ Your device has **low VRAM** and cannot load full-precision models.  
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.  

📌 **Avoid Quantized Models if:**  
❌ You need **maximum accuracy** (full-precision models are better for this).  
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).  

---

### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**  
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.  

- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.  
  - **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.  
  - **Trade-off**: Lower accuracy compared to higher-bit quantizations.  

- **IQ3_S**: Small block size for **maximum memory efficiency**.  
  - **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.  

- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.  
  - **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.  

- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.  
  - **Use case**: Best for **low-memory devices** where **Q6_K** is too large.  

- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.  
  - **Use case**: Best for **ARM-based devices** or **low-memory environments**.  

---

### **Summary Table: Model Format Selection**  

| Model Format  | Precision  | Memory Usage  | Device Requirements  | Best Use Case  |  
|--------------|------------|---------------|----------------------|---------------|  
| **BF16**     | Highest    | High          | BF16-supported GPU/CPUs  | High-speed inference with reduced memory |  
| **F16**      | High       | High          | FP16-supported devices | GPU inference when BF16 isn't available |  
| **Q4_K**     | Medium Low | Low           | CPU or Low-VRAM devices | Best for memory-constrained environments |  
| **Q6_K**     | Medium     | Moderate      | CPU with more memory | Better accuracy while still being quantized |  
| **Q8_0**     | High       | Moderate      | CPU or GPU with enough VRAM | Best accuracy among quantized models |  
| **IQ3_XS**   | Very Low   | Very Low      | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |  
| **Q4_0**     | Low        | Low           | ARM or low-memory devices | llama.cpp can optimize for ARM devices |  

---

## **Included Files & Details**  

### `QwQ-32B-bf16.gguf`  
- Model weights preserved in **BF16**.  
- Use this if you want to **requantize** the model into a different format.  
- Best if your device supports **BF16 acceleration**.  

### `QwQ-32B-f16.gguf`  
- Model weights stored in **F16**.  
- Use if your device supports **FP16**, especially if BF16 is not available.  

### `QwQ-32B-bf16-q8_0.gguf`  
- **Output & embeddings** remain in **BF16**.  
- All other layers quantized to **Q8_0**.  
- Use if your device supports **BF16** and you want a quantized version.  

### `QwQ-32B-f16-q8_0.gguf`  
- **Output & embeddings** remain in **F16**.  
- All other layers quantized to **Q8_0**.    

### `QwQ-32B-q4_k.gguf`  
- **Output & embeddings** quantized to **Q8_0**.  
- All other layers quantized to **Q4_K**.  
- Good for **CPU inference** with limited memory.  

### `QwQ-32B-q4_k_s.gguf`  
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.  
- Best for **very low-memory setups**.  

### `QwQ-32B-q6_k.gguf`  
- **Output & embeddings** quantized to **Q8_0**.  
- All other layers quantized to **Q6_K** .  

### `QwQ-32B-q8_0.gguf`  
- Fully **Q8** quantized model for better accuracy.  
- Requires **more memory** but offers higher precision.  

### `QwQ-32B-iq3_xs.gguf`  
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.  
- Best for **ultra-low-memory devices**.  

### `QwQ-32B-iq3_m.gguf`  
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.  
- Suitable for **low-memory devices**.  

### `QwQ-32B-q4_0.gguf`  
- Pure **Q4_0** quantization, optimized for **ARM devices**.  
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.

# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**  
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:  
👉 [Free Network Monitor](https://freenetworkmonitor.click/dashboard)  

💬 **How to test**:  
1. Click the **chat icon** (bottom right on any page)  
2. Choose an **AI assistant type**:  
   - `TurboLLM` (GPT-4-mini)  
   - `FreeLLM` (Open-source)  
   - `TestLLM` (Experimental CPU-only)  

### **What I’m Testing**  
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:  
- **Function calling** against live network services  
- **How small can a model go** while still handling:  
  - Automated **Nmap scans**  
  - **Quantum-readiness checks**  
  - **Metasploit integration**  

🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):  
- ✅ **Zero-configuration setup**  
- ⏳ 30s load time (slow inference but **no API costs**)  
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!  

### **Other Assistants**  
🟢 **TurboLLM** – Uses **gpt-4-mini** for:  
- **Real-time network diagnostics**  
- **Automated penetration testing** (Nmap/Metasploit)  
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://freenetworkmonitor.click/download)  

🔵 **HugLLM** – Open-source models (≈8B params):  
- **2x more tokens** than TurboLLM  
- **AI-powered log analysis**  
- 🌐 Runs on Hugging Face Inference API  

### 💡 **Example AI Commands to Test**:  
1. `"Give me info on my websites SSL certificate"`  
2. `"Check if my server is using quantum safe encyption for communication"`  
3. `"Run a quick Nmap vulnerability test"`  



# QwQ-32B

<a href="https://chat.qwenlm.ai/" target="_blank" style="margin: 2px;">
    <img alt="Chat" src="https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5" style="display: inline-block; vertical-align: middle;"/>
</a>

## Introduction

QwQ is the reasoning model of the Qwen series. Compared with conventional instruction-tuned models, QwQ, which is capable of thinking and reasoning, can achieve significantly enhanced performance in downstream tasks, especially hard problems. QwQ-32B is the medium-sized reasoning model, which is capable of achieving competitive performance against state-of-the-art reasoning models, e.g., DeepSeek-R1, o1-mini.

<p align="center">
  <img width="100%" src="figures/benchmark.jpg">
</p>


**This repo contains the QwQ 32B model**, which has the following features:
- Type: Causal Language Models
- Training Stage: Pretraining & Post-training (Supervised Finetuning and Reinforcement Learning)
- Architecture: transformers with RoPE, SwiGLU, RMSNorm, and Attention QKV bias
- Number of Parameters: 32.5B
- Number of Paramaters (Non-Embedding): 31.0B
- Number of Layers: 64
- Number of Attention Heads (GQA): 40 for Q and 8 for KV
- Context Length: Full 131,072 tokens
    - For prompts exceeding 8,192 tokens in length, you must enable YaRN as outlined in [this section](#usage-guidelines).

**Note:** For the best experience, please review the [usage guidelines](#usage-guidelines) before deploying QwQ models.

You can try our [demo](https://huggingface.co/spaces/Qwen/QwQ-32B-Demo) or access QwQ models via [QwenChat](https://chat.qwen.ai).

For more details, please refer to our [blog](https://qwenlm.github.io/blog/qwq-32b/), [GitHub](https://github.com/QwenLM/Qwen2.5), and [Documentation](https://qwen.readthedocs.io/en/latest/).

## Requirements

QwQ is based on Qwen2.5, whose code has been in the latest Hugging face `transformers`. We advise you to use the latest version of `transformers`.

With `transformers<4.37.0`, you will encounter the following error:
```
KeyError: 'qwen2'
```

## Quickstart

Here provides a code snippet with `apply_chat_template` to show you how to load the tokenizer and model and how to generate contents.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Qwen/QwQ-32B"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "How many r's are in the word \"strawberry\""
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)

model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
```

### Usage Guidelines

To achieve optimal performance, we recommend the following settings:

1. **Enforce Thoughtful Output**: Ensure the model starts with "\<think\>\n" to prevent generating empty thinking content, which can degrade output quality. If you use `apply_chat_template` and set `add_generation_prompt=True`, this is already automatically implemented, but it may cause the response to lack the \<think\> tag at the beginning. This is normal behavior.

2. **Sampling Parameters**:
   - Use Temperature=0.6, TopP=0.95, MinP=0 instead of Greedy decoding to avoid endless repetitions.
   - Use TopK between 20 and 40 to filter out rare token occurrences while maintaining the diversity of the generated output.
   - For supported frameworks, you can adjust the `presence_penalty` parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may result in occasional language mixing and a slight decrease in performance.

3. **No Thinking Content in History**: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. This feature is already implemented in `apply_chat_template`.

4. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
   - **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
   - **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the `answer` field with only the choice letter, e.g.,`\"answer\": \"C\"`." in the prompt.

5. **Handle Long Inputs**: For inputs exceeding 8,192 tokens, enable [YaRN](https://arxiv.org/abs/2309.00071) to improve the model's ability to capture long-sequence information effectively.

    For supported frameworks, you could add the following to `config.json` to enable YaRN:
    ```json
    {
    ...,
    "rope_scaling": {
        "factor": 4.0,
        "original_max_position_embeddings": 32768,
        "type": "yarn"
    }
    }
    ```

    For deployment, we recommend using vLLM. Please refer to our [Documentation](https://qwen.readthedocs.io/en/latest/deployment/vllm.html) for usage if you are not familar with vLLM.
    Presently, vLLM only supports static YARN, which means the scaling factor remains constant regardless of input length, **potentially impacting performance on shorter texts**. 
    We advise adding the `rope_scaling` configuration only when processing long contexts is required.

## Evaluation & Performance

Detailed evaluation results are reported in this [📑 blog](https://qwenlm.github.io/blog/qwq-32b/).

For requirements on GPU memory and the respective throughput, see results [here](https://qwen.readthedocs.io/en/latest/benchmark/speed_benchmark.html).

## Citation

If you find our work helpful, feel free to give us a cite.

```
@misc{qwq32b,
    title = {QwQ-32B: Embracing the Power of Reinforcement Learning},
    url = {https://qwenlm.github.io/blog/qwq-32b/},
    author = {Qwen Team},
    month = {March},
    year = {2025}
}

@article{qwen2.5,
      title={Qwen2.5 Technical Report}, 
      author={An Yang and Baosong Yang and Beichen Zhang and Binyuan Hui and Bo Zheng and Bowen Yu and Chengyuan Li and Dayiheng Liu and Fei Huang and Haoran Wei and Huan Lin and Jian Yang and Jianhong Tu and Jianwei Zhang and Jianxin Yang and Jiaxi Yang and Jingren Zhou and Junyang Lin and Kai Dang and Keming Lu and Keqin Bao and Kexin Yang and Le Yu and Mei Li and Mingfeng Xue and Pei Zhang and Qin Zhu and Rui Men and Runji Lin and Tianhao Li and Tianyi Tang and Tingyu Xia and Xingzhang Ren and Xuancheng Ren and Yang Fan and Yang Su and Yichang Zhang and Yu Wan and Yuqiong Liu and Zeyu Cui and Zhenru Zhang and Zihan Qiu},
      journal={arXiv preprint arXiv:2412.15115},
      year={2024}
}
```