Instructions to use qwfasf/K2-Horizon-7B-UnrealEngine with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use qwfasf/K2-Horizon-7B-UnrealEngine with Transformers:

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

pipe = pipeline("text-generation", model="qwfasf/K2-Horizon-7B-UnrealEngine", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("qwfasf/K2-Horizon-7B-UnrealEngine", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use qwfasf/K2-Horizon-7B-UnrealEngine 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 qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf qwfasf/K2-Horizon-7B-UnrealEngine: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 qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf qwfasf/K2-Horizon-7B-UnrealEngine: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 qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M

Use Docker

docker model run hf.co/qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M

LM Studio
Jan

vLLM

How to use qwfasf/K2-Horizon-7B-UnrealEngine with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "qwfasf/K2-Horizon-7B-UnrealEngine"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "qwfasf/K2-Horizon-7B-UnrealEngine",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M

SGLang

How to use qwfasf/K2-Horizon-7B-UnrealEngine 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 "qwfasf/K2-Horizon-7B-UnrealEngine" \
    --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": "qwfasf/K2-Horizon-7B-UnrealEngine",
		"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 "qwfasf/K2-Horizon-7B-UnrealEngine" \
        --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": "qwfasf/K2-Horizon-7B-UnrealEngine",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use qwfasf/K2-Horizon-7B-UnrealEngine with Ollama:
```
ollama run hf.co/qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M
```
Unsloth Desktop

How to use qwfasf/K2-Horizon-7B-UnrealEngine with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use qwfasf/K2-Horizon-7B-UnrealEngine with Docker Model Runner:
```
docker model run hf.co/qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M
```

Lemonade

How to use qwfasf/K2-Horizon-7B-UnrealEngine with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M

Run and chat with the model

lemonade run user.K2-Horizon-7B-UnrealEngine-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use qwfasf/K2-Horizon-7B-UnrealEngine with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf qwfasf/K2-Horizon-7B-UnrealEngine: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 qwfasf/K2-Horizon-7B-UnrealEngine:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use qwfasf/K2-Horizon-7B-UnrealEngine with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf qwfasf/K2-Horizon-7B-UnrealEngine: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 "qwfasf/K2-Horizon-7B-UnrealEngine: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"

K2-Horizon-7B Unreal Engine (Blueprint + C++)

This repository contains a standalone BF16 merge of the K2-Horizon-7B base model and the v152 Unreal Engine Blueprint + C++ adapter. The adapter was applied once to the original, unquantized BF16 base. This is export revision v160.

The model is intended for Unreal Engine Blueprint and C++ assistance. The export itself has not been evaluated with generation benchmarks; please validate its behavior on your own tasks.

GGUF for 8 GB GPUs

Use K2-Horizon-7B-UnrealEngine-Q4_K_M.gguf for local inference on an 8 GB GPU. It is a 5.59 GB Q4_K_M quantization of the merged BF16 model. The original BF16 Transformers weights remain available in this repository.

Recommended LM Studio starting settings:

Context length: 8,192 tokens
GPU offload: maximum/all layers
Flash Attention: enabled
If LM Studio reports insufficient VRAM, lower the context length to 4,096 tokens

The model metadata advertises a 524,288-token maximum context, but that context size is not practical on an 8 GB GPU. Quantization reduces model weight memory; KV cache memory still grows with context length.

In LM Studio, download or import the Q4_K_M GGUF file rather than the BF16 GGUF or Safetensors files.

Model details

Base model: IFM/K2-Horizon-7B
Adapter: v152 Blueprint + C++ adapter
Export: standalone BF16 weights, adapter merged once
Architecture: K2HorizonForCausalLM (custom Transformers implementation included in this repository)
Transformers version used for the export: 5.17.0
Position configuration: max_position_embeddings = 524288
Tokenizer and chat template are included
Sharded model files total 18,019,196,335 bytes (~18.0 GB); runtime memory use is higher and depends on device and context length

The 524,288 position setting is a configuration value, not a claim that long-context quality has been evaluated for this export.

Load with Transformers

Install PyTorch, Accelerate, and the specified Transformers version in an environment suitable for BF16 inference:

pip install "transformers==5.17.0" accelerate

The repository includes custom model code. trust_remote_code=True allows Transformers to load and execute that code; review the files before enabling it if you do not trust the repository source.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "qwfasf/K2-Horizon-7B-UnrealEngine"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=True,
    dtype=torch.bfloat16,
    device_map="auto",
)

The export manifest records that the standalone model reload without the adapter passed. Generation and benchmark evaluation were not run for this export.

Merge and validation notes

The v152 adapter was merged once into the original unquantized BF16 base; no optimizer or training updates were performed during export.
The adapter was trained using NF4, while this export merges it into the unquantized BF16 base. BF16 merge rounding means this export may differ numerically from live NF4 + LoRA inference.
Export provenance reports 252 merged modules, 504 adapter tensors checked, and 327 saved tensors verified against the merged model.
The tokenizer vocabulary and sample template were checked against the parent model.
Adapter SHA-256: 62ff493e8cae7c706a9eda313ada15e7baeb04a51c11f651650bd8c4f202f363
See merge_provenance.json and export_manifest.json for the recorded checks and file hashes.

License

This derivative is published under the Apache-2.0 license, following the base model card. Review the base model's license and terms before use.

Downloads last month: 122

Safetensors

Model size

9B params

Tensor type

BF16

Model tree for qwfasf/K2-Horizon-7B-UnrealEngine

Base model

IFM/K2-Horizon-7B

Quantized

(46)

this model

Quantizations

2 models