Instructions to use IFM/K2-Think-V2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use IFM/K2-Think-V2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="IFM/K2-Think-V2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("IFM/K2-Think-V2")
model = AutoModelForCausalLM.from_pretrained("IFM/K2-Think-V2", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use IFM/K2-Think-V2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "IFM/K2-Think-V2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "IFM/K2-Think-V2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/IFM/K2-Think-V2

SGLang

How to use IFM/K2-Think-V2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "IFM/K2-Think-V2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "IFM/K2-Think-V2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "IFM/K2-Think-V2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "IFM/K2-Think-V2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use IFM/K2-Think-V2 with Docker Model Runner:
```
docker model run hf.co/IFM/K2-Think-V2
```

twkillian commited on Jan 13

Commit

ebd296f

verified ·

1 Parent(s): 0387738

Revised K2 Think naming

Browse files

Files changed (1) hide show

README.md +11 -11

README.md CHANGED Viewed

@@ -8,7 +8,7 @@ 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)
@@ -16,7 +16,7 @@ pipeline_tag: text-generation
 <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
@@ -36,7 +36,7 @@ We use the following serving configurations:
 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.
@@ -44,7 +44,7 @@ The chat template is directly inherited from K2-V2-Instruct, with the default `r
 from transformers import pipeline
 import torch
-model_id =  "LLM360/K2-Think-70B"
 pipe = pipeline(
     "text-generation",
@@ -75,7 +75,7 @@ client = OpenAI(
 )
 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]"}
@@ -93,7 +93,7 @@ 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 |
@@ -105,7 +105,7 @@ A summary of evaluation results are reported in our [Blog]()
 <!-- ## 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) |
 | --------------------------------- | ----------------------: | ---------------------------------: |
@@ -144,12 +144,12 @@ We have employed various techniques to reduce bias, harmful outputs, and other r
 ---
 # 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},
 }

 pipeline_tag: text-generation
 ---
+# K2 Think (Jan '26): A Fully-Sovereign Reasoning System
 📚 [Paper]() - 📝 [Code](https://github.com/LLM360/Reasoning360) - 🏢 [Project Page](https://k2think.ai)
 <br>
+K2 Think (Jan '26') 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
 The provided chat template sets the reasoning effort to `high`
 ### Transformers
+You can use `K2 Think (Jan '26)` 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.
 from transformers import pipeline
 import torch
+model_id =  "LLM360/K2-Think-0126"
 pipe = pipeline(
     "text-generation",
 )
 completion = client.chat.completions.create(
+    model="LLM360/K2-Think-0126",
     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]"}
 ## Benchmarks (pass\@1, average over 16 runs)
+| Domain  | Benchmark        | K2 Think (Jan '26) |
 | ------- | --------------------  | -----------: |
 | Math    | AIME 2025             |        90.42 |
 | Math    | HMMT 2025             |        84.79 |
 <!-- ## Inference Speed
+We deploy K2 THINK (Jan '26) 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) |
 | --------------------------------- | ----------------------: | ---------------------------------: |
 ---
 # Citation
+If you use K2 Think (Jan '26) in your research, please use the following citation:
 ```bibtex
+@misc{k2thinkteam2026k2think0126,
+      title={K2 {T}hink ({Jan} '26): 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},
 }