File size: 5,802 Bytes
af1dd50 b5255e1 af1dd50 b5255e1 af1dd50 b14e9cd 4656a5b b14e9cd af1dd50 b5255e1 af1dd50 b14e9cd af1dd50 b5255e1 4538eb5 b5255e1 af1dd50 b5255e1 af1dd50 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 | ---
base_model:
- LLM360/K2-V2-Instruct
language:
- en
library_name: transformers
license: apache-2.0
pipeline_tag: text-generation
---
# K2-Think (70B): A Fully-Sovereign Reasoning System
📚 [Paper]() - 📝 [Code](https://github.com/LLM360/Reasoning360) - 🏢 [Project Page](https://k2think.ai)
<center><img src="banner.png" alt="k2-think-banner"/></center>
<br>
K2-Think (70B) is a 70 billion parameter open-weights general reasoning model with strong performance in competitive mathematical problem solving built on-top of [K2-V2-Instruct](huggingface.co/LLM360/K2-V2-Instruct), comprising a fully sovereign reasoning system.
# Quickstart
### Serving configurations
We use the following serving configurations:
| Setting | Value |
| - | - |
| Temperature | 1.0 |
| Top-p | 1.0 |
| Top-k | -1 |
| Context Length | 131072 |
| Context Length Extension | 2x using YaRN |
| Chat Template | Default provided in `chat_template.jinja` |
The provided chat template sets the reasoning effort to `high`
### Transformers
You can use `K2-Think (70B)` with Transformers. If you use `transformers.pipeline`, it will apply the chat template automatically. If you use `model.generate` directly, you need to apply the chat template mannually.
The chat template is directly inherited from K2-V2-Instruct, with the default `reasoning_effort` set to `"high"`. The other levels of reasoning effort (`"low"` and `"medium"`) are still available but have not been tested or evaluated. As such, the model's behavior under such settings is not assured to maintain reported performance.
```python
from transformers import pipeline
import torch
model_id = "LLM360/K2-Think-70B"
pipe = pipeline(
"text-generation",
model=model_id,
torch_dtype="auto",
device_map="auto",
)
messages = [
{"role": "user", "content": "what is the next prime number after 2600?"},
]
outputs = pipe(
messages,
max_new_tokens=131072,
)
print(outputs[0]["generated_text"][-1])
```
If you cannot use `tokenizer.apply_chat_template`, you may also pass in these arguments using `extra_body` and `chat_template_kwargs`:
```
from openai import OpenAI
client = OpenAI(
base_url="http://localhost:8000/v1",
api_key="key"
)
completion = client.chat.completions.create(
model="LLM360/K2-Think-70B",
messages = [
{"role": "system", "content": "You are K2-Think, a helpful assistant created by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) Institute of Foundation Models (IFM)."},
{"role": "user", "content": "Solve the 24 game [2, 3, 5, 6]"}
],
extra_body={
"chat_template_kwargs": {"reasoning_effort": "high"},
},
)
```
---
# Evaluation & Performance
A summary of evaluation results are reported in our [Blog]()
## Benchmarks (pass\@1, average over 16 runs)
| Domain | Benchmark | K2-Think (70B) |
| ------- | -------------------- | -----------: |
| Math | AIME 2025 | 90.42 |
| Math | HMMT 2025 | 84.79 |
| Code | LiveCodeBench v5 | TBD |
| Science | GPQA-Diamond | 72.98 |
| Science | Humanity's Last Exam | TBD |
<!-- --- -->
<!-- ## Inference Speed
We deploy K2-THINK (70B) on Cerebras Wafer-Scale Engine (WSE) systems, leveraging the world’s largest processor and speculative decoding to achieve unprecedented inference speeds for our 32B reasoning system.
| Platform | Throughput (tokens/sec) | Example: 32k-token response (time) |
| --------------------------------- | ----------------------: | ---------------------------------: |
| **Cerebras WSE (our deployment)** | **\~2,000** | **\~16 s** |
| Typical Cloud Service setup | \~200 | \~160 s | -->
<!-- --- -->
<!-- ## Safety Evaluation
Aggregated across four safety dimensions (**Safety-4**):
| Aspect | Macro-Avg |
| ------------------------------- | --------: |
| High-Risk Content Refusal | 0.83 |
| Conversational Robustness | 0.89 |
| Cybersecurity & Data Protection | 0.56 |
| Jailbreak Resistance | 0.72 |
| **Safety-4 Macro (avg)** | **0.75** |
--- -->
# Terms of Use
We have employed various techniques to reduce bias, harmful outputs, and other risks in the model. While these efforts help improve safety and reliability, the model, like all Large Language Models, may still generate inaccurate, misleading, biased, or otherwise undesirable content. By downloading, using, or interacting with this model, you acknowledge these limitations and agree to the following:
1. **Prohibited Uses**
- You may **not** use this model for any **illegal, unlawful, or harmful activities**, including but not limited to fraud, abuse, harassment, privacy violations, or the creation/dissemination of malicious content.
2. **User Responsibility**
- You are solely responsible for how you use the model and for any outcomes that result from its use.
- The authors and institutions involved in releasing this model do **not** accept liability for any consequences arising from its use.
3. **No Warranty**
- The model is provided **“as is” without any warranties or guarantees**.
---
# Citation
If you use K2-Think (70B) in your research, please use the following citation:
```bibtex
@misc{k2thinkteam2026k2think70B,
title={K2-{T}hink 70{B}: A Fully-Sovereign Reasoning System},
author={K2-Think Team and Taylor W. Killian and Varad Pimpalkhute and Richard Fan and Haonan Li and Chengqian Gao and Ming Shan Hee and Xudong Han and John Maggs and Guowei He and Zhengzhong Liu and Eric P. Xing},
year={2026},
url={https://tbd.org},
}
``` |