Instructions to use gbuzhf/KAT-Coder-V2.5-Dev-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 gbuzhf/KAT-Coder-V2.5-Dev-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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./llama-cli -hf gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

LM Studio
Jan
Ollama
How to use gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF with Ollama:
```
ollama run hf.co/gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL
```

Unsloth Studio

How to use gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF 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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF 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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF to start chatting

How to use gbuzhf/KAT-Coder-V2.5-Dev-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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

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": "gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use gbuzhf/KAT-Coder-V2.5-Dev-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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use gbuzhf/KAT-Coder-V2.5-Dev-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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

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 "gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL" \
  --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 gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF with Docker Model Runner:
```
docker model run hf.co/gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL
```

Lemonade

How to use gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull gbuzhf/KAT-Coder-V2.5-Dev-MTP-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.KAT-Coder-V2.5-Dev-MTP-GGUF-UD-Q4_K_XL

List all available models

lemonade list

This is the final Repo.

pinned

by gbuzhf - opened 1 day ago

Discussion

gbuzhf

Owner 1 day ago

•

edited about 11 hours ago

Original MTP grafted into KAT quantizations -> Bartowski's imatrix -> results were fair.

I then started native-MTP training, and the measurements were promising: 30%+ performance increase, with a native imatrix blended into Bartowski's.
I deleted the first repo and kept that one while I ran real-world tests.

In real-world measurements (real harness, temp environment), the native-MTP experiment failed against the original MTP.

The native-MTP repo is now deleted. Back to: original MTP grafted into KAT quantizations -> blended imatrix. This is the final repo and it will be kept.

-- My reasoning was:

I had been following the Kimi K2 -> K3 and MiniMax experiment papers.

Basically:

Qwen 3.6 35B-A3B = 1.0
Original MTP     = 1.0
KAT Coder V2.5 Dev = 1.2 (fine-tuned Qwen)
So if we fine-tune the original MTP on KAT's outputs and hidden states, it should land at 1.2, and we'd recover the original performance there.

Synthetic benchmarks worked.

Real world: failed.

Therefore: experiment documented and removed.

gbuzhf pinned discussion 1 day ago

gbuzhf locked this discussion 1 day ago

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