Instructions to use kalomaze/Mistral-7b-MoEified-8x with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use kalomaze/Mistral-7b-MoEified-8x with Transformers:

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

pipe = pipeline("text-generation", model="kalomaze/Mistral-7b-MoEified-8x")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("kalomaze/Mistral-7b-MoEified-8x")
model = AutoModelForCausalLM.from_pretrained("kalomaze/Mistral-7b-MoEified-8x")

llama-cpp-python

How to use kalomaze/Mistral-7b-MoEified-8x with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="kalomaze/Mistral-7b-MoEified-8x",
	filename="ggml-model-f16.gguf",
)

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

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use kalomaze/Mistral-7b-MoEified-8x with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf kalomaze/Mistral-7b-MoEified-8x:F16
# Run inference directly in the terminal:
llama-cli -hf kalomaze/Mistral-7b-MoEified-8x:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf kalomaze/Mistral-7b-MoEified-8x:F16
# Run inference directly in the terminal:
llama-cli -hf kalomaze/Mistral-7b-MoEified-8x:F16

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 kalomaze/Mistral-7b-MoEified-8x:F16
# Run inference directly in the terminal:
./llama-cli -hf kalomaze/Mistral-7b-MoEified-8x:F16

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 kalomaze/Mistral-7b-MoEified-8x:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf kalomaze/Mistral-7b-MoEified-8x:F16

Use Docker

docker model run hf.co/kalomaze/Mistral-7b-MoEified-8x:F16

LM Studio
Jan

vLLM

How to use kalomaze/Mistral-7b-MoEified-8x with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "kalomaze/Mistral-7b-MoEified-8x"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kalomaze/Mistral-7b-MoEified-8x",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/kalomaze/Mistral-7b-MoEified-8x:F16

SGLang

How to use kalomaze/Mistral-7b-MoEified-8x 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 "kalomaze/Mistral-7b-MoEified-8x" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kalomaze/Mistral-7b-MoEified-8x",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "kalomaze/Mistral-7b-MoEified-8x" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kalomaze/Mistral-7b-MoEified-8x",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Ollama
How to use kalomaze/Mistral-7b-MoEified-8x with Ollama:
```
ollama run hf.co/kalomaze/Mistral-7b-MoEified-8x:F16
```

Unsloth Studio new

How to use kalomaze/Mistral-7b-MoEified-8x 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 kalomaze/Mistral-7b-MoEified-8x 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 kalomaze/Mistral-7b-MoEified-8x to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for kalomaze/Mistral-7b-MoEified-8x to start chatting

Docker Model Runner
How to use kalomaze/Mistral-7b-MoEified-8x with Docker Model Runner:
```
docker model run hf.co/kalomaze/Mistral-7b-MoEified-8x:F16
```

Lemonade

How to use kalomaze/Mistral-7b-MoEified-8x with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull kalomaze/Mistral-7b-MoEified-8x:F16

Run and chat with the model

lemonade run user.Mistral-7b-MoEified-8x-F16

List all available models

lemonade list

Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Wtf is a MoEification?!

Turns out, you can slice up the individual MLP layers of a dense language model into even splits of experts.

What I did here:

Split the MLP projections (gate, down, proj) into the amount of total experts you want (in this case, I just went with 8 experts).
Multiply the values of the parameters for the down-projection by the total amount of experts (so the magnitude of the activation outputs, when averaged linearly together, ends up being equivalent)
Initialize the router layers with zeroes, so the expert usage is completely equal by default and has no unintentional biases as a consequence of random initialization being done the normal way.

As a result, the model behaves completely coherently when all 8 experts are activated (i.e, experts_per_tok is equal to 8.)

With 4 experts activated, it's... far less coherent.

Ok but why?

I am interested in the prospect of continuing to train this in such a way where it can naturally handle variable expert counts, and learn to balance the features. If this works, we can potentially teach the behavior of using less computation for tokens that are trivial to predict, while using more when necessary.

Also thanks StefanGliga for giving me the idea while we were discussing this paper :3

Downloads last month: 602

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for kalomaze/Mistral-7b-MoEified-8x

Quantizations

1 model

Paper for kalomaze/Mistral-7b-MoEified-8x

Sparse MoE as the New Dropout: Scaling Dense and Self-Slimmable Transformers

Paper • 2303.01610 • Published Mar 2, 2023