moonshotai
/

Kimi-K2-Thinking

Text Generation

compressed-tensors

Model card Files Files and versions

Instructions to use moonshotai/Kimi-K2-Thinking with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use moonshotai/Kimi-K2-Thinking with Transformers:

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

pipe = pipeline("text-generation", model="moonshotai/Kimi-K2-Thinking", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("moonshotai/Kimi-K2-Thinking", trust_remote_code=True, dtype="auto")

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps

How to use moonshotai/Kimi-K2-Thinking with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/moonshotai/Kimi-K2-Thinking

How to use moonshotai/Kimi-K2-Thinking 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 "moonshotai/Kimi-K2-Thinking" \
    --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": "moonshotai/Kimi-K2-Thinking",
		"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 "moonshotai/Kimi-K2-Thinking" \
        --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": "moonshotai/Kimi-K2-Thinking",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use moonshotai/Kimi-K2-Thinking with Docker Model Runner:
```
docker model run hf.co/moonshotai/Kimi-K2-Thinking
```

Resources

View closed (14)

Add SWE-bench Verified evaluation result (71.3%)

#60 opened 2 months ago by

Install & run moonshotai/Kimi-K2-Thinking easily using llmpm

#58 opened 2 months ago by

Fine tuning kimik2

#57 opened 3 months ago by

savsuresharintra

Add Terminal-Bench evaluation result (35.7%)

#56 opened 3 months ago by

Add converted tokenizer (no trust_remote_code needed)

#55 opened 3 months ago by

Add HLE evaluation result (with tools)

#54 opened 4 months ago by

Add GPQA evaluation result

#52 opened 4 months ago by

I think `eos_token_id` should be `163585`in `config.json`?

#51 opened 4 months ago by

Add community evaluation results for GPQA

#50 opened 4 months ago by

Add MMLU-Pro evaluation result

#49 opened 4 months ago by

Add HLE evaluation result

#48 opened 4 months ago by

New tool to benchmark GGUF files

#47 opened 5 months ago by

This model thinks that repeating itself is good

#46 opened 5 months ago by

Which Longform Writing benchmark is?

#44 opened 5 months ago by

Update tokenization_kimi.py

#43 opened 5 months ago by

need the version of transformer

#42 opened 5 months ago by

Kimi-K2-Thinking vLLM self host tool call fail

#40 opened 5 months ago by

I need a teacher

#39 opened 5 months ago by

How to limit the thinking duration?

#38 opened 6 months ago by

Add Artificial Analysis evaluations for kimi-k2-thinking

#37 opened 6 months ago by

mackenzietechdocs

tokenizer.apply_chat_template is right? Manually Parsing Tool Calls Failed! but openai api Success, Why?

#36 opened 6 months ago by

Kimi-K2 (non-thinking) QAT

#35 opened 6 months ago by

Token Count Calculation in SFT Data Distribution Curation

#31 opened 6 months ago by

kimi-k2-thinking can't use a tool with 2 required arguments if it has 2 or more tools.

#30 opened 6 months ago by

Is the reasoning parser going to be available in vllm soon?

#29 opened 6 months ago by

dingobingobango

If we apply PTQ to a QAT model, what will happen

#28 opened 6 months ago by

User Reviews about K2 Thinking and Search Agent

#27 opened 6 months ago by

Was this really trained during QAT using a symmetric 4bit quant with only 15/16 values used?

#26 opened 6 months ago by

Their API Takes Your Data - PRIVACY RISK

#24 opened 6 months ago by

Any plan to open source the search agent framework?

#22 opened 6 months ago by

Beware quantization destroyed coding abilities, broken code in Q5-K-M (735Gb RAM)

#21 opened 6 months ago by

Does the Ktranformers deployment support non-AVX-512?

#16 opened 6 months ago by

Question: Will models with other precisions be published soon?

#14 opened 6 months ago by

Question: What model precision was used during the benchmarks?

#13 opened 6 months ago by

vllm deployment failed

#11 opened 6 months ago by

How to control reasoning effort? Heavy Mode/low-latency mode?

#10 opened 6 months ago by

huggingfacemotnt

MoEQuant?

#7 opened 6 months ago by

Did you set GPT-5's reasoning effort to high?

#6 opened 6 months ago by

K2 Thinking Browsecomp/HLE Reproducibility | 结果复现

#5 opened 6 months ago by

Awesome work! Do you want to try AMO-Bench, the most challenging MO-level benchmark?

#3 opened 6 months ago by

Python script to decompress tensors?

#2 opened 7 months ago by

Video of Step-by-Step Review and Testing

#1 opened 7 months ago by