Instructions to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-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 notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF: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 notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF: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 notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M

Use Docker

docker model run hf.co/notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M

Ollama
How to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF with Ollama:
```
ollama run hf.co/notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M
```
Unsloth Desktop

How to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF: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": "notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF with Docker Model Runner:
```
docker model run hf.co/notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M
```

Lemonade

How to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Step-3.7-Flash-Q4_K_M-MTP-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF: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 notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF: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 "notSnix/Step-3.7-Flash-Q4_K_M-MTP-GGUF: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"

Difficulty loading due to the naming of layers (`blk.45` vs `blk.0`)

by iandennismiller - opened Jun 1

Discussion

iandennismiller

Jun 1

Very interested to try out the MTP draft! However, I was unable to run using the llama.cpp commit you referenced. It seems like llama.cpp wanted a draft model that started with layer 0, but you've named it after the layer where it belongs when MTP is merged.

Would it work to simply rename the layers in the draft models?

notSnix

Owner Jun 1

The uploaded draft files intentionally keep the original tail-layer
names (blk.45-blk.47) because the tested llama.cpp Step MTP loader treats
the draft as the MTP tail of a 48-layer model (nextn_predict_layers=3), not as
a normal standalone draft model.

If llama.cpp is asking for blk.0, it is likely using the normal speculative
draft path instead of the Step MTP path. make sure the command includes
--spec-type draft-mtp and is built from the referenced/current commit. A simple
tensor rename is probably not gonna work, because the metadata still says
block_count=48 and nextn_predict_layers=3. This is just how the Step model is

notSnix

Owner Jun 1

Or just ask codex/claude to figure it out and fix it

notSnix

Owner Jun 1

ive updated the repo with better instructions

iandennismiller

Jun 1

Amazing. I did attempt to rebuild the MTP draft model using the gguf python library but it was not as simple as renaming the layers; the geometry was wrong. Thanks for taking a look.

iandennismiller

Jun 1

Okay, so that patch is non-trivial. I've watched how PRs go with llama.cpp and I suspect this isn't getting merged without some feedback from the core devs. (I'm not one).

This is really cool work, though ... have you already got a PR going for this?

iandennismiller

Jun 2

FYI, this PR was just merged and your MTP models work just fine now when compiling the master llama.cpp branch.

https://github.com/ggml-org/llama.cpp/pull/23274

notSnix changed discussion status to closed Jun 3

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