Model Overview
Model Features
Model Capabilities
Use Cases
๐ QwQ-32B GGUF Models
QwQ-32B GGUF models offer ultra-low-bit quantization solutions, enhancing memory efficiency without significant loss of accuracy. These models are suitable for various hardware environments and use cases, from high-performance inference to resource-constrained devices.
โจ Features
Ultra-Low-Bit Quantization with IQ-DynamicGate (1 - 2 bit)
- Precision - Adaptive Quantization: Our latest method introduces precision - adaptive quantization for ultra - low - bit models (1 - 2 bit). Benchmark tests on Llama - 3 - 8B show significant improvements.
- Layer - Specific Strategies: It uses layer - specific strategies to preserve accuracy while maintaining extreme memory efficiency.
Benchmark Context
- Testing Setup: All tests were conducted on Llama - 3 - 8B - Instruct using a standard perplexity evaluation pipeline, a 2048 - token context window, and the same prompt set across all quantizations.
Method
- Dynamic Precision Allocation:
- First/Last 25% of layers โ IQ4_XS (selected layers)
- Middle 50% โ IQ2_XXS/IQ3_S (increase efficiency)
- Critical Component Protection:
- Embeddings/output layers use Q5_K, reducing error propagation by 38% compared to standard 1 - 2 bit quantization.
Quantization Performance Comparison (Llama - 3 - 8B)
Quantization | Standard PPL | DynamicGate PPL | ฮ PPL | Std Size | DG Size | ฮ 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 Improvements:
- ๐ฅ IQ1_M shows a massive 43.9% perplexity reduction (from 27.46 to 15.41).
- ๐ IQ2_S cuts perplexity by 36.9% while adding only 0.2GB.
- โก IQ1_S maintains 39.7% better accuracy despite 1 - bit quantization.
Trade - offs:
- All variants have modest size increases (0.1 - 0.3GB).
- Inference speeds remain comparable (<5% difference).
When to Use These Models
- ๐ Fitting models into GPU VRAM
- โ Memory - constrained deployments
- โ CPU and Edge Devices where 1 - 2 bit errors can be tolerated
- โ Research into ultra - low - bit quantization
๐ฆ Installation
No installation steps are provided in the original document.
๐ป Usage Examples
Choosing the Right Model Format
Selecting the correct model format depends on your hardware capabilities and memory constraints.
BF16 (Brain Float 16) โ Use if BF16 acceleration is available
- Features: A 16 - bit floating - point format designed for faster computation while retaining good precision. It provides a similar dynamic range as FP32 but with lower memory usage.
- Use Cases:
- ๐ Use BF16 if:
- โ Your hardware has native BF16 support (e.g., newer GPUs, TPUs).
- โ You want higher precision while saving memory.
- โ You plan to requantize the model into another format.
- ๐ Avoid BF16 if:
- โ Your hardware does not support BF16 (it may fall back to FP32 and run slower).
- โ You need compatibility with older devices that lack BF16 optimization.
- ๐ Use BF16 if:
F16 (Float 16) โ More widely supported than BF16
- Features: A 16 - bit floating - point format with high precision but a smaller range of values than BF16. It works on most devices with FP16 acceleration support (including many GPUs and some CPUs).
- Use Cases:
- ๐ Use F16 if:
- โ Your hardware supports FP16 but not BF16.
- โ You need a balance between speed, memory usage, and accuracy.
- โ You are running on a GPU or another device optimized for FP16 computations.
- ๐ Avoid F16 if:
- โ Your device lacks native FP16 support (it may run slower than expected).
- โ You have memory limitations.
- ๐ Use F16 if:
Quantized Models (Q4_K, Q6_K, Q8, etc.) โ For CPU & Low - VRAM Inference
- Features: Quantization reduces model size and memory usage while maintaining as much accuracy as possible. Lower - bit models (Q4_K) are best for minimal memory usage but may have lower precision, while higher - bit models (Q6_K, Q8_0) offer better accuracy but require more memory.
- Use Cases:
- ๐ Use Quantized Models if:
- โ You are running inference on a CPU and need an optimized model.
- โ Your device has low VRAM and cannot load full - precision models.
- โ You want to reduce memory footprint while keeping reasonable accuracy.
- ๐ Avoid Quantized Models if:
- โ You need maximum accuracy (full - precision models are better for this).
- โ Your hardware has enough VRAM for higher - precision formats (BF16/F16).
- ๐ Use Quantized Models if:
Very Low - Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)
- IQ3_XS: Ultra - low - bit quantization (3 - bit) with extreme memory efficiency. Best for ultra - low - memory devices where even Q4_K is too large. However, it has lower accuracy compared to higher - bit quantizations.
- IQ3_S: Small block size for maximum memory efficiency. Best for low - memory devices where IQ3_XS is too aggressive.
- IQ3_M: Medium block size for better accuracy than IQ3_S. Suitable for low - memory devices where IQ3_S is too limiting.
- Q4_K: 4 - bit quantization with block - wise optimization for better accuracy. Best for low - memory devices where Q6_K is too large.
- Q4_0: Pure 4 - bit quantization, optimized for ARM devices. 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 are 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 are 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, and all other layers are 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, and all other layers are quantized to Q8_0.QwQ-32B-q4_k.gguf
: Output & embeddings are quantized to Q8_0, and all other layers are quantized to Q4_K. Good for CPU inference with limited memory.QwQ-32B-q4_k_s.gguf
: The 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 are quantized to Q8_0, and all other layers are quantized to Q6_K.QwQ-32B-q8_0.gguf
: A 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.
Quickstart
Here is a code snippet with apply_chat_template
to show you how to load the tokenizer and model and how to generate contents.
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:
- 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 setadd_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. - 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.
- 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
. - 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.
- Handle Long Inputs: For inputs exceeding 8,192 tokens, enable YaRN 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:
For deployment, we recommend using vLLM. Please refer to our Documentation for usage if you are not familiar 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": { "factor": 4.0, "original_max_position_embeddings": 32768, "type": "yarn" } }
rope_scaling
configuration only when processing long contexts is required.
๐ Documentation
Evaluation & Performance
Detailed evaluation results are reported in this ๐ blog. For requirements on GPU memory and the respective throughput, see results here.
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}
}
๐ง Technical Details
No technical details beyond what is already covered in the features and usage sections are provided in the original document.
๐ License
This project is licensed under the Apache - 2.0 License.
๐ If you find these models useful
โค Please click "Like" if you find this useful!
Help me test my AI - Powered Network Monitor Assistant with quantum - ready security checks:
๐ Free Network Monitor
๐ฌ How to test:
- Click the chat icon (bottom right on any page)
- Choose an AI assistant type:
TurboLLM
(GPT - 4 - mini)FreeLLM
(Open - source)TestLLM
(Experimental CPU - only)
What Iโm Testing
Iโm 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
๐ก TestLLM โ Current experimental model (llama.cpp on 6 CPU threads):
- โ Zero - configuration setup
- โณ 30s load time (slow inference but no API costs)
- ๐ง Help wanted! If youโre into edge - device AI, letโs collaborate!
Other Assistants
๐ข TurboLLM โ Uses gpt - 4 - mini for:
- Real - time network diagnostics
- Automated penetration testing (Nmap/Metasploit)
- ๐ Get more tokens by downloading our Free Network Monitor Agent
๐ต HugLLM โ Open - source models (โ8B params):
- 2x more tokens than TurboLLM
- AI - powered log analysis
- ๐ Runs on Hugging Face Inference API
๐ก Example AI Commands to Test
"Give me info on my websites SSL certificate"
"Check if my server is using quantum safe encyption for communication"
"Run a quick Nmap vulnerability test"

