---
library_name: transformers
license: apache-2.0
base_model: Qwen/Qwen2.5-7B-Instruct
tags:
- llama-factory
- full
- generated_from_trainer
model-index:
- name: OpenThinker3-7B
  results: []
datasets:
- open-thoughts/OpenThoughts3-1.2M
---

# <span style="color: #7FFF7F;">OpenThinker3-7B GGUF Models</span>


## <span style="color: #7F7FFF;">Model Generation Details</span>

This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`745aa531`](https://github.com/ggerganov/llama.cpp/commit/745aa5319b9930068aff5e87cf5e9eef7227339b).




## <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 |  

---



# <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://readyforquantum.com/dashboard/?assistant=open)  

💬 **How to test**:  
 Choose an **AI assistant type**:  
   - `TurboLLM` (GPT-4o-mini)  
   - `HugLLM` (Hugginface 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**  
  - **Network Monitoring tasks**  

🟡 **TestLLM** – Current experimental model (llama.cpp on 2 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-4o-mini** for: 
- **Create custom cmd processors to run .net code on Free Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)  
- 🔑 Get more tokens by logging in or [downloading our Free Network Monitor Agent with integrated AI Assistant](https://readyforquantum.com/download)  

🔵 **HugLLM** – Latest Open-source models:  
- 🌐 Runs on Hugging Face Inference API  

### 💡 **Example commands to you could 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 comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Free Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!





<p align="center">
    <img src="https://huggingface.co/datasets/open-thoughts/open-thoughts-114k/resolve/main/open_thoughts.png" width="50%">
</p>

<p align="center">
<a href="https://arxiv.org/abs/2506.04178" style="margin-right: 24px;">paper</a> |
<a href="https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M" style="margin-right: 24px; margin-left: 24px;">dataset</a> |
<a href="https://huggingface.co/open-thoughts/OpenThinker3-7B" style="margin-left: 24px;">model</a>
</p>

> [!NOTE]
> We have released a paper for OpenThoughts! See our paper [here](https://arxiv.org/abs/2506.04178).

# OpenThinker3-7B

State-of-the-art open-data 7B reasoning model. 🚀

This model is a fine-tuned version of [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) on the 
[OpenThoughts3-1.2M](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M) dataset. 
It represents a notable improvement over our previous models, [OpenThinker-7B](https://huggingface.co/open-thoughts/OpenThinker-7B) and [OpenThinker2-7B](https://huggingface.co/open-thoughts/OpenThinker2-7B), and it outperforms several other strong reasoning 7B models such as [DeepSeek-R1-Distill-Qwen-7B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B) and [Llama-3.1-Nemotron-Nano-8B-v1](https://huggingface.co/nvidia/Llama-3.1-Nemotron-Nano-8B-v1), despite being trained only with SFT, without any RL.

This time, we also released a paper! See our [paper](https://arxiv.org/abs/2506.04178) and [blog post](https://openthoughts.ai/blog/ot3) for more details. OpenThinker3-32B to follow! 👀

# Evaluation Results
The numbers reported in the table below are evaluated with our open-source tool [Evalchemy](https://github.com/mlfoundations/Evalchemy). 
In the table below, we bold values in each column that are within 2 standard errors of the best.

| Model                                                                                           | Data  | AIME24 | AIME25 |  AMC23 | MATH500 | HMMT O2/25 | LCB 06/24-01/25 | CodeElo | CodeForces | GPQA-D | JEEBench |
| ----------------------------------------------------------------------------------------------- | ----- | ------ | ------ | ------ | ------- | ---------- | --------------- | ------- | ---------- | ------ | -------- |
| [OpenThinker-7B](https://huggingface.co/open-thoughts/OpenThinker-7B)                           | ✅    |  30.7  |  22.0  |  72.5  |   82.8  |   15.7     |    26.1         |  11.1   |  14.9      |  38.6  |  45.3    |
| [OpenThinker2-7B](https://huggingface.co/open-thoughts/OpenThinker2-7B)                         | ✅    |  60.7  |  38.7  |  89.8  |   87.6  |   24.7     |    40.6         |  22.8   |  26.6      |  47.0  |  65.1    |
| **[OpenThinker3-7B](https://huggingface.co/open-thoughts/OpenThinker3-7B)**                     | ✅    |**69.0**|**53.3**|**93.5**| **90.0**|   **42.7** |    **51.7**     |  31.0   |**32.2**    |  53.7  |**72.4**  |
| [DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) | ❌    |  51.3  |  38.0  |  92.0  |   88.0  |   25.0     |    34.5         |  19.9   |  21.1      |  33.2  |  50.4    |
| [OpenR1-Distill-7B](https://huggingface.co/open-r1/OpenR1-Distill-7B)                           | ✅    |  57.7  |  39.7  |  87.0  |   88.0  |   25.7     |    30.7         |  30.1   |  29.3      |**58.9**|  68.7    |
| [Llama-3.1-Nemotron-Nano-8B-v1](https://huggingface.co/nvidia/Llama-3.1-Nemotron-Nano-8B-v1)    | ✅    |  62.0  |  48.0  |**94.0**|   89.4  |   26.7     |    **50.9**     |  30.9   |**32.9**    |  52.9  |  70.7    |
| [AceReason-Nemotron-7B](https://huggingface.co/nvidia/AceReason-Nemotron-7B)                    | ✅    |**71.0**|  50.7  |**93.8**|   89.8  |   33.3     |    44.3         |**32.9** |**30.9**    |  52.9  |  64.3    |

# Data

This model was trained on the [OpenThoughts3-1.2M](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M) dataset.

The key to the strong model performance is our comprehensive data pipeline and over 1,000+ ablation experiments.
This led to the creation of [OpenThoughts3-1.2M](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M), which consists of 850,000 math questions, 250,000 code questions, and 100,000 science questions.
Reasoning traces are generated with QwQ-32B. 

See the [OpenThoughts3-1.2M](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M) dataset page or our [paper](https://arxiv.org/abs/2506.04178) for additional information.


# Intended uses & limitations

Apache 2.0 License


## Training procedure

We used 512 A100 nodes to train the model for 48 hours.

## Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 8e-05
- seed: 42
- distributed_type: multi-GPU
- num_devices: 512
- gradient_accumulation_steps: 1
- total_train_batch_size: 512
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 5.0
- weight_decay: 0.0

## Framework versions

- Transformers 4.46.1
- Pytorch 2.3.0
- Datasets 3.1.0
- Tokenizers 0.20.3

More info can be found in our repository: [https://github.com/open-thoughts/open-thoughts](https://github.com/open-thoughts/open-thoughts).

# Links
- 📝 [OpenThoughts Paper](https://arxiv.org/abs/2506.04178)
- 📊 [OpenThoughts3-1.2M and OpenThinker3-7B Blog Post](https://www.open-thoughts.ai/blog/ot3)
- 💻 [Open Thoughts GitHub Repository](https://github.com/open-thoughts/open-thoughts)
- 🧠 [OpenThoughts3-1.2M dataset](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M)
- 🤖 [OpenThinker3-7B model](https://huggingface.co/open-thoughts/OpenThinker3-7B) - this model.

# Citation
```
@misc{guha2025openthoughtsdatarecipesreasoning,
  title={OpenThoughts: Data Recipes for Reasoning Models}, 
  author={Etash Guha and Ryan Marten and Sedrick Keh and Negin Raoof and Georgios Smyrnis and Hritik Bansal and Marianna Nezhurina and Jean Mercat and Trung Vu and Zayne Sprague and Ashima Suvarna and Benjamin Feuer and Liangyu Chen and Zaid Khan and Eric Frankel and Sachin Grover and Caroline Choi and Niklas Muennighoff and Shiye Su and Wanjia Zhao and John Yang and Shreyas Pimpalgaonkar and Kartik Sharma and Charlie Cheng-Jie Ji and Yichuan Deng and Sarah Pratt and Vivek Ramanujan and Jon Saad-Falcon and Jeffrey Li and Achal Dave and Alon Albalak and Kushal Arora and Blake Wulfe and Chinmay Hegde and Greg Durrett and Sewoong Oh and Mohit Bansal and Saadia Gabriel and Aditya Grover and Kai-Wei Chang and Vaishaal Shankar and Aaron Gokaslan and Mike A. Merrill and Tatsunori Hashimoto and Yejin Choi and Jenia Jitsev and Reinhard Heckel and Maheswaran Sathiamoorthy and Alexandros G. Dimakis and Ludwig Schmidt},
  year={2025},
  eprint={2506.04178},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2506.04178}, 
}
```