Instructions to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit",
	filename="K-EXAONE-236B-A23B-Q4_K_M-00001-of-00003.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit: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 mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit: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 mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M

Use Docker

docker model run hf.co/mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M

LM Studio
Jan

vLLM

How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M

Ollama
How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with Ollama:
```
ollama run hf.co/mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M
```

Unsloth Studio new

How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit 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 mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit 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 mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit to start chatting

Pi new

How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit: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": "mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit: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 mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with Docker Model Runner:
```
docker model run hf.co/mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M
```

Lemonade

How to use mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull mkreynolds/K-EXAONE-236B-A23B-GGUF-4bit:Q4_K_M

Run and chat with the model

lemonade run user.K-EXAONE-236B-A23B-GGUF-4bit-Q4_K_M

List all available models

lemonade list

high hallucinations

by krustik - opened Jan 7

Discussion

krustik

Jan 7

I've tried it with that special branch of llama and got very high hallucinations with code test. It just stuck in reasoning forever loop trying to determine the correct notes melody.
I will quantize myself the original model into Q8 version and try that, but 96 model shards are not so user friendly to download, it fact the main problem to get it from huggingface at all, git in xet or lfs (no matter) clonning they disconnect for some reason all the time.
Q4 uses ~150Gb RAM

krustik

Jan 12

I've checked it in Q8 (uses ~255Gb RAM) and F16(original, uses ~474Gb RAM).
Unfortunately it's useless for any coding, same self looping even in original size, also kinda repeating thoughts many times in reasoning over and over. It basically gives up to make code for me on reasoning stage, which is very weird to see in model of such size.
Example:

However, if the user is interested in a simplified example, perhaps taking a small excerpt from a symphony and converting that into ChucK code would be more feasible.

For example, let's take the first few bars of the first movement of Mozart's Symphony No. 40 (K. 550). Then, we can write a ChucK program that plays those few bars using a piano instrument.

But even then, the user might not be aware that writing out the entire symphony would require an enormous amount of code, which is not practical to do manually.

So, perhaps the best approach is to provide a ChucK code example that plays a small excerpt from a Mozart symphony, using a piano instrument, and explain that converting the entire symphony into ChucK code is not feasible due to the sheer length and complexity.

But the user specifically said "in its full length", so they want the entire symphony. However, given that it's impossible to do manually, perhaps the user is mistaken.

Alternatively, maybe the user is asking for a ChucK program that uses a piano instrument to play the symphony, but they don't realize that the code would need to be written out note by note.

Given all this, perhaps the most appropriate response is to explain that converting a full-length Mozart symphony into ChucK code using only a piano instrument is not feasible due to the sheer length and complexity of the symphony.

However, if the user is interested in a simplified example, perhaps taking a small excerpt from a symphony and converting that into ChucK code would be more feasible.

So, for example, let's take the first four bars of the first movement of Mozart's Symphony No. 40 (K. 550), and convert that into ChucK code.

But even then,

Also i've checked it for literature, also quite weak.
Haven't checked on censorship and anything else, dont have time on this, because new Deepseek downloaded and waiting.
Someone maybe find the use for this model.

krustik changed discussion status to closed Jan 12

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