unsloth
/

gemma-4-12b-it-GGUF

Image-Text-to-Text

Model card Files Files and versions

Instructions to use unsloth/gemma-4-12b-it-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 unsloth/gemma-4-12b-it-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 unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/gemma-4-12b-it-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 unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/gemma-4-12b-it-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 unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./llama-cli -hf unsloth/gemma-4-12b-it-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 unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL

LM Studio
Jan

How to use unsloth/gemma-4-12b-it-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "unsloth/gemma-4-12b-it-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": "unsloth/gemma-4-12b-it-GGUF",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL

Ollama
How to use unsloth/gemma-4-12b-it-GGUF with Ollama:
```
ollama run hf.co/unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL
```
Unsloth Desktop

How to use unsloth/gemma-4-12b-it-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL

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": "unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use unsloth/gemma-4-12b-it-GGUF with Docker Model Runner:
```
docker model run hf.co/unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL
```

How to use unsloth/gemma-4-12b-it-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.gemma-4-12b-it-GGUF-UD-Q4_K_XL

List all available models

lemonade list

How to use unsloth/gemma-4-12b-it-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 unsloth/gemma-4-12b-it-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 unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Atomic Chat

How to use unsloth/gemma-4-12b-it-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/gemma-4-12b-it-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 "unsloth/gemma-4-12b-it-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"

Error downloading MTP model

#18

by Magnarks - opened Jun 10

Jun 10

I'm trying to download the MTP model using llama.cpp, but I'm getting this error:
Downloading mtp-gemma-4-12b-it.gguf 0%[K[1B
Downloading mmproj-BF16.gguf ── 6%[K[1B
[34m0.07.224.090[0m [31mE common_download_file_single_online: unable to rename file: C:\Users\diego.cache\huggingface\hub\models--unsloth--gemma-4-12b-it-GGUF\blobs\145db9094bc0f85f1701e255a2ed216dcc9800fc8bc8631ad00905b456bd451b.downloadInProgress to C:\Users\diego.cache\huggingface\hub\models--unsloth--gemma-4-12b-it-GGUF\blobs\145db9094bc0f85f1701e255a2ed216dcc9800fc8bc8631ad00905b456bd451b
[0m[34m0.07.224.096[0m [31mE common_download_file_single_online: download failed after 3 attempts
failed to download model from Hugging Face────────────────────────── 100%[K[3B

Jun 10

If I try to use the normal version, it downloads the MTP but still gives an error.

InformaticsSolutions

Jun 10

+1.
I manually downloaded the mtp gguf:

hf download unsloth/gemma-4-12b-it-GGUF --local-dir /home/user1/models/gemma4-12b --include "MTP/*Q8_0*"

and tried loading it with:

llama-server -m ${models}/gemma4-12b/gemma-4-12b-it-UD-Q8_K_XL.gguf -fit off -c 16384 -ngl 999 -ub 1024 -b 512 --temp 1 --min-p 0.1 --top-p 0.95 --top-k 64 --no-mmap -np 1  --model-draft ${models}/gemma4-12b/MTP/gemma-4-12b-it-Q8_0-MTP.gguf --spec-type draft-mtp --spec-draft-n-max 3 --verbose

Getting:

[...]
llama-swap-1  | 0.09.613.462 D sched_reserve: worst-case: n_tokens = 512, n_seqs = 1, n_outputs = 1
llama-swap-1  | 0.09.613.466 D graph_reserve: reserving a graph for ubatch with n_tokens =    1, n_seqs =  1, n_outputs =    1
llama-swap-1  | 0.09.614.479 I sched_reserve: Flash Attention was auto, set to enabled
llama-swap-1  | 0.09.614.496 I sched_reserve: resolving fused Gated Delta Net support:
llama-swap-1  | 0.09.614.498 D graph_reserve: reserving a graph for ubatch with n_tokens =    1, n_seqs =  1, n_outputs =    1
llama-swap-1  | 0.09.615.415 I sched_reserve: fused Gated Delta Net (autoregressive) enabled
llama-swap-1  | 0.09.615.437 D graph_reserve: reserving a graph for ubatch with n_tokens =   16, n_seqs =  1, n_outputs =   16
llama-swap-1  | 0.09.616.391 I sched_reserve: fused Gated Delta Net (chunked) enabled
llama-swap-1  | 0.09.616.415 D graph_reserve: reserving a graph for ubatch with n_tokens =  512, n_seqs =  1, n_outputs =    4
llama-swap-1  | 0.09.692.762 D ggml_cuda_host_malloc: failed to allocate 32.52 MiB of pinned memory: out of memory
llama-swap-1  | 0.09.692.925 D graph_reserve: reserving a graph for ubatch with n_tokens =    1, n_seqs =  1, n_outputs =    1
llama-swap-1  | 0.09.694.765 D graph_reserve: reserving a graph for ubatch with n_tokens =  512, n_seqs =  1, n_outputs =    4
llama-swap-1  | 0.09.696.225 I sched_reserve:      CUDA0 compute buffer size =   139.52 MiB
llama-swap-1  | 0.09.696.250 I sched_reserve:  CUDA_Host compute buffer size =    32.52 MiB
[...]
llama-swap-1  | 0.10.149.165 I llama_prepare_model_devices: using device CUDA0 (NVIDIA GeForce RTX 3080 Laptop GPU) (0000:01:00.0) - 0 MiB free
[...]
llama-swap-1  | 0.10.699.523 E llama_model_load: error loading model: vector::_M_range_check: __n (which is 1) >= this->size() (which is 1)
llama-swap-1  | 0.10.699.559 E llama_model_load_from_file_impl: failed to load model
llama-swap-1  | 0.10.699.564 E srv    load_model: failed to load draft model, '/models/gemma4-12b/MTP/gemma-4-12b-it-Q8_0-MTP.gguf'
llama-swap-1  | 0.10.699.575 I srv    operator(): operator(): cleaning up before exit...
llama-swap-1  | 0.10.701.546 E srv  llama_server: exiting due to model loading error
llama-swap-1  | 0.10.701.672 D ~llama_context:      CUDA0 compute buffer size is 139.5177 MiB, matches expectation of 139.5177 MiB
llama-swap-1  | 0.10.701.674 D ~llama_context:  CUDA_Host compute buffer size is  32.5196 MiB, matches expectation of  32.5196 MiB
[...]

./llama-server --version
version: 9590 (d2462f8f7)
built with GNU 11.4.0 for Linux x86_64

Nvidia RTX 3080 t1, 16Gb VRAM

InformaticsSolutions

Jun 11

Please disregard my message above: turns out there was something using VRAM that i was not aware of. After restarting MTP works correctly. Getting 44tps vs 25 without MTP. Nice. Thanks you!

Jun 12

Please disregard my message above: turns out there was something using VRAM that i was not aware of. After restarting MTP works correctly. Getting 44tps vs 25 without MTP. Nice. Thanks you!

Thanks anyway, I managed to get it working, although I'm not seeing the same improvement in T/S as you at the moment.

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