Instructions to use AtomicChat/Kimi-K3-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AtomicChat/Kimi-K3-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="AtomicChat/Kimi-K3-GGUF",
	filename="Q2_K/Kimi-K3-Q2_K-00001-of-00024.gguf",
)

output = llm(
	"Once upon a time,",
	max_tokens=512,
	echo=True
)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use AtomicChat/Kimi-K3-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf AtomicChat/Kimi-K3-GGUF:Q2_K
# Run inference directly in the terminal:
llama cli -hf AtomicChat/Kimi-K3-GGUF:Q2_K

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf AtomicChat/Kimi-K3-GGUF:Q2_K
# Run inference directly in the terminal:
llama cli -hf AtomicChat/Kimi-K3-GGUF:Q2_K

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf AtomicChat/Kimi-K3-GGUF:Q2_K
# Run inference directly in the terminal:
./llama-cli -hf AtomicChat/Kimi-K3-GGUF:Q2_K

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf AtomicChat/Kimi-K3-GGUF:Q2_K
# Run inference directly in the terminal:
./build/bin/llama-cli -hf AtomicChat/Kimi-K3-GGUF:Q2_K

Use Docker

docker model run hf.co/AtomicChat/Kimi-K3-GGUF:Q2_K

LM Studio
Jan

vLLM

How to use AtomicChat/Kimi-K3-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AtomicChat/Kimi-K3-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AtomicChat/Kimi-K3-GGUF",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/AtomicChat/Kimi-K3-GGUF:Q2_K

Ollama
How to use AtomicChat/Kimi-K3-GGUF with Ollama:
```
ollama run hf.co/AtomicChat/Kimi-K3-GGUF:Q2_K
```

Unsloth Studio

How to use AtomicChat/Kimi-K3-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for AtomicChat/Kimi-K3-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for AtomicChat/Kimi-K3-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for AtomicChat/Kimi-K3-GGUF to start chatting

Atomic Chat new
Docker Model Runner
How to use AtomicChat/Kimi-K3-GGUF with Docker Model Runner:
```
docker model run hf.co/AtomicChat/Kimi-K3-GGUF:Q2_K
```

Lemonade

How to use AtomicChat/Kimi-K3-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull AtomicChat/Kimi-K3-GGUF:Q2_K

Run and chat with the model

lemonade run user.Kimi-K3-GGUF-Q2_K

List all available models

lemonade list

Kimi-K3-GGUF

GGUF quantizations of moonshotai/Kimi-K3 (mirrored at AtomicChat/Kimi-K3).

Kimi K3 is a ~2.9T-parameter (104B active) mixture-of-experts model: 93 layers, 896 experts (16 active per token), 7168 hidden, 1M context, natively trained in MXFP4. The original ships only in MXFP4, so everything here derives from a direct MXFP4 to Q8_0 conversion; lower-bit quants are requantized from that Q8_0 base.

Read this before downloading: sizes

This is a massive model. Even the smallest 1-bit quant is ~540 GB — larger than the 512 GB that all-but-the-biggest machines top out at. To run a quant you need enough fast memory (RAM + VRAM combined) to hold the whole file plus overhead. This repo targets large-RAM servers and multi-GPU rigs, not desktops. Pick the largest quant that fits your machine's memory.

Requirements

kimi_k3 is not yet in mainline llama.cpp. These files need a build with Kimi K3 support:

Fork: AtomicBot-ai/atomic-llama-cpp-turboquant, branch feat/kimi-k3-support

Mainline llama.cpp rejects these with an unknown-architecture error until support is merged upstream (tracking: ggml-org/llama.cpp#26185).

Available quants

Each quant is a folder of 45 GB shards; point llama.cpp at the first shard (...-00001-of-000NN.gguf), the rest load automatically. Sizes for not-yet-uploaded quants are estimates.

Folder	Bits	Size	imatrix	Status	Notes
`IQ1_S`	1.56	~540 GB	yes	building	Smallest; extreme, quality-degraded
`IQ1_M`	1.75	~610 GB	yes	planned
`IQ2_XXS`	2.06	~715 GB	yes	planned	Best size/quality in the 2-bit tier
`Q2_K`	2.9	~1009 GB	no	uploaded
`Q3_K_S`	3.4	~1196 GB	no	uploading
`Q8_0`	8.5	~2955 GB	no	uploaded	Conversion base; for archival / requant, not for running
`imatrix.dat`	-	small	-	uploaded	Importance matrix (see below)

We deliberately skip the mid-size i-quants (IQ2_XS/IQ2_S/IQ3/IQ4): at ~0.8-1.2 TB they overlap with Q2_K/Q3_K_S, which already cover that range more simply. The set above spans the useful spectrum without redundancy.

imatrix

imatrix.dat was computed with the calibration_datav3 corpus and is applied to every i-quant (IQ1/IQ2).

Honest notes on how it was made:

Source: the Q3_K_S quant, not the full Q8_0. Kimi K3 multi-GPU inference does not yet work in the fork, and the full 3 TB Q8 does not fit in RAM for CPU inference; Q3_K_S (~1.2 TB) fits and runs cleanly on CPU. imatrix computed from a quantized model is standard practice; the effect on the final 1-2 bit outputs is second-order relative to the quantization itself.
Coverage of the 896 experts is partial by nature (only 16 fire per token); rarely-activated experts fall back to default quantization for their weights.

Usage

# download one quant folder
hf download AtomicChat/Kimi-K3-GGUF --include "IQ2_XXS/*" --local-dir ./kimi-k3

# run with the fork build
./llama-cli -m ./kimi-k3/IQ2_XXS/Kimi-K3-IQ2_XXS-00001-of-000NN.gguf -p "Hello"

Attribution and license

Derivative of Kimi K3 by Moonshot AI, redistributed under the Kimi K3 License; the original copyright and license are preserved in LICENSE.

Kimi K3 License
Copyright (c) 2026 Moonshot AI

Quantized and published by AtomicChat.

Downloads last month: -

GGUF

Model size

2.8T params

Architecture

kimi-k3

Hardware compatibility

2-bit

3-bit

8-bit

Model tree for AtomicChat/Kimi-K3-GGUF

Base model

moonshotai/Kimi-K3

Quantized

(14)

this model