Instructions to use moonshotai/Kimi-K2.6 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use moonshotai/Kimi-K2.6 with Transformers:

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

pipe = pipeline("image-text-to-text", model="moonshotai/Kimi-K2.6", trust_remote_code=True)
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("moonshotai/Kimi-K2.6", trust_remote_code=True, dtype="auto")

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use moonshotai/Kimi-K2.6 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "moonshotai/Kimi-K2.6"
# 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-K2.6",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/moonshotai/Kimi-K2.6

SGLang

How to use moonshotai/Kimi-K2.6 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-K2.6" \
    --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-K2.6",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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-K2.6" \
        --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-K2.6",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use moonshotai/Kimi-K2.6 with Docker Model Runner:
```
docker model run hf.co/moonshotai/Kimi-K2.6
```

fix: add missing use_deterministic_attn parameter to MoonViT3dEncoder

#22

by ace-coreweave - opened 13 days ago

base: refs/heads/main

←

from: refs/pr/22

Discussion Files changed

+19

-4

ace-coreweave

13 days ago

MoonViT3dEncoder.init references self.use_deterministic_attn on line 575
when constructing the MoonViTEncoderLayer blocks, but the attribute is never
set on self. Loading the model via AutoModelForCausalLM with
trust_remote_code=True raises:

AttributeError: 'MoonViT3dEncoder' object has no attribute
                'use_deterministic_attn'

The sibling class MoonViTEncoderLayer already accepts use_deterministic_attn
as a keyword parameter with default False, so the attribute on the parent
3d-encoder was clearly intended to plumb through the same flag. Restore the
missing parameter with the same default.

Production serving paths (vLLM's Kimi-K25 model executor) bypass the HF
custom modeling init and construct the vision tower differently, so this
bug is invisible at serving time but blocks transformers-based workflows
like ModelOpt NVFP4 quantization and HF-native fine-tuning.

Identical fix already merged in Kimi-K2.5 PR #91 (by @katuni4ka , approved
by @fxmarty-amd ). This mirrors it to K2.6 byte-for-byte.

Minimal repro:

from transformers import AutoModelForCausalLM
AutoModelForCausalLM.from_pretrained(
    "moonshotai/Kimi-K2.6", trust_remote_code=True, torch_dtype="auto",
)

fix: add missing use_deterministic_attn parameter to MoonViT3dEncoderf13e4d07

bdellabe

8 days ago

This resolved my issue, @bigeagle can we merge this in so we don't require other users to manually update?

bigeagle

8 days ago

thanks for your contribution!

also fix weight init issuebed55d49

bigmoyan

Moonshot AI org 8 days ago

@bdellabe @ace-coreweave Hi, I've also added some code to fix the weight initialization issue. AutoModelForCausalLM.from_pretrained now works on my end. However, this doesn't mean transformers inference is fully supported — if you plan to implement Kimi k2.6 inference in other frameworks, please mainly refer to the vLLM/SGLang implementation.

bigmoyan changed pull request status to merged 8 days ago

bigmoyan

Moonshot AI org 7 days ago

@ace-coreweave @bdellabe see: https://huggingface.co/moonshotai/Kimi-K2.6/discussions/26

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment