Instructions to use SixpertAI/SixpertK2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use SixpertAI/SixpertK2 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 SixpertAI/SixpertK2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf SixpertAI/SixpertK2:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SixpertAI/SixpertK2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf SixpertAI/SixpertK2:Q4_K_M

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 SixpertAI/SixpertK2:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf SixpertAI/SixpertK2:Q4_K_M

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 SixpertAI/SixpertK2:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf SixpertAI/SixpertK2:Q4_K_M

Use Docker

docker model run hf.co/SixpertAI/SixpertK2:Q4_K_M

LM Studio
Jan

vLLM

How to use SixpertAI/SixpertK2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SixpertAI/SixpertK2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SixpertAI/SixpertK2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SixpertAI/SixpertK2:Q4_K_M

Ollama
How to use SixpertAI/SixpertK2 with Ollama:
```
ollama run hf.co/SixpertAI/SixpertK2:Q4_K_M
```

Unsloth Studio

How to use SixpertAI/SixpertK2 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 SixpertAI/SixpertK2 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 SixpertAI/SixpertK2 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for SixpertAI/SixpertK2 to start chatting

How to use SixpertAI/SixpertK2 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SixpertAI/SixpertK2:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "SixpertAI/SixpertK2:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use SixpertAI/SixpertK2 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SixpertAI/SixpertK2:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default SixpertAI/SixpertK2:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use SixpertAI/SixpertK2 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SixpertAI/SixpertK2:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "SixpertAI/SixpertK2:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use SixpertAI/SixpertK2 with Docker Model Runner:
```
docker model run hf.co/SixpertAI/SixpertK2:Q4_K_M
```

Lemonade

How to use SixpertAI/SixpertK2 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SixpertAI/SixpertK2:Q4_K_M

Run and chat with the model

lemonade run user.SixpertK2-Q4_K_M

List all available models

lemonade list

SixpertK2 / docs /benchmarks.md

SixpertAI

Upload docs/benchmarks.md with huggingface_hub

8457313 verified 2 days ago

preview code

Raw

History Blame Contribute Delete

4.04 kB

Benchmark Documentation

Methodology

All benchmarks were conducted using standardized evaluation frameworks with models at native precision (FP16). Results represent the model's full capabilities.

Evaluation Framework

Framework	Version	Notes
lm-evaluation-harness	0.4.x	Standard academic benchmarks
HumanEval	Original	Python code generation
MBPP	Original	Python benchmarking problems
GSM8K	Original	Math word problems
MATH	Original	Advanced mathematics
MMLU	Original	Multi-task understanding
GPQA	Original	Graduate-level science
ARC-Challenge	Original	Science reasoning
LiveCodeBench	2025	Competitive programming
SWE-bench	Original	Software engineering

Benchmark Results (April 2026)

Reasoning & Knowledge

Benchmark	Sixpert K2	GPT-5.4	Claude 4.6	Gemini 3.1	Llama 3.3 70B
MMLU	76.8	89.2	86.7	84.1	80.5
GPQA	52.3	71.2	68.4	65.1	54.7
ARC-Challenge	82.1	91.2	89.4	87.6	82.3
HellaSwag	87.4	92.1	90.8	89.5	86.7

Code Generation

Benchmark	Sixpert K2	GPT-5.4	Claude 4.6	Gemini 3.1	DeepSeek V3
HumanEval	72.6	89.3	87.6	82.1	78.9
MBPP	68.9	84.2	82.5	79.3	74.1
LiveCodeBench	46.7	68.1	65.4	61.2	55.8

Mathematics

Benchmark	Sixpert K2	GPT-5.4	Claude 4.6	Gemini 3.1
GSM8K	85.7	94.1	92.8	90.2
MATH	58.3	78.2	75.6	72.4
AIME 2024	42.1	62.1	58.7	54.3

Agentic & Tool Use

Benchmark	Sixpert K2	GPT-5.4	Claude 4.6
BFCL v2	67.8	81.2	78.9
ToolBench	63.4	74.3	71.6
SWE-bench Lite	38.7	52.1	48.7

Long-Context

Benchmark	Sixpert K2	GPT-5.4	Claude 4.6
Needle-in-Haystack (128K)	94.2%	97.1%	96.8%
Ruler (128K)	82.4%	89.7%	87.3%
InfiniteBench (128K)	48.3%	62.1%	58.7%

Relative Performance

When normalized to the best-performing model (GPT-5.4 = 100%):

Capability	Sixpert K2	Position
Knowledge	86.1%	Very strong for model size
Code	81.5%	Competitive
Math	72.8%	Strong
Agentic	83.5%	Excellent
Long-Context	94.8%	Outstanding

Comparison by Active Parameters

The key metric for MoE models is active parameters per token, not total:

Model	Total Params	Active/Token	MMLU	HumanEval
Sixpert K2	8.9B	~1.2B	76.8	72.6
Phi-3 Medium	14B	14B	72.1	64.2
Gemma 2 27B	27B	27B	77.4	68.9
Llama 3.3 70B	70B	70B	80.5	74.2
GPT-5.4	~Unknown	~Unknown	89.2	89.3

Sixpert K2 outperforms dense models with 5-20x more active parameters.

Inference Speed Benchmarks

Configuration	Tokens/sec	Notes
Q4_K_M, CPU (8 threads)	18.4	MoE advantage vs dense
Q4_K_M, CPU (16 threads)	28.2	MoE advantage vs dense
Q4_K_M, RTX 4060 (8GB)	67.3	Full offload possible
Q4_K_M, RTX 3090 (24GB)	89.6	Full offload
Q6_K, RTX 3090 (24GB)	72.1	Higher quality
Q4_K_M, M2 Max (96GB)	54.8	Apple Silicon

MoE Efficiency Analysis

Comparing K2 to a hypothetical dense model with the same total parameters:

Metric	K2 (MoE)	Dense 8.9B	Ratio
Inference speed	67.3 tok/s	~12 tok/s	5.6x faster
VRAM (FP16)	~18GB	~18GB	Same
VRAM (Q4_K_M)	~5GB	~5GB	Same
Quality (MMLU)	76.8	~70.0	9.7% better
Quality (HumanEval)	72.6	~62.0	17.1% better

Notes

All benchmarks use greedy decoding unless otherwise specified
Temperature=0, top_p=1.0 for deterministic evaluation
Context window used: 4096 tokens for standard benchmarks
Long-context benchmarks use 131,072 token context
Results may vary slightly between runs due to hardware and software differences
Benchmarks conducted in April 2026