Instructions to use OnlyTextLLMs/gemma-4-12B-it-OnlyText-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 OnlyTextLLMs/gemma-4-12B-it-OnlyText-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 OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf OnlyTextLLMs/gemma-4-12B-it-OnlyText-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 OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf OnlyTextLLMs/gemma-4-12B-it-OnlyText-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 OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf OnlyTextLLMs/gemma-4-12B-it-OnlyText-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 OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M

Use Docker

docker model run hf.co/OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OnlyTextLLMs/gemma-4-12B-it-OnlyText-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": "OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M

Ollama
How to use OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF with Ollama:
```
ollama run hf.co/OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M
```
Unsloth Desktop

How to use OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf OnlyTextLLMs/gemma-4-12B-it-OnlyText-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": "OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF with Docker Model Runner:
```
docker model run hf.co/OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M
```

Lemonade

How to use OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.gemma-4-12B-it-OnlyText-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use OnlyTextLLMs/gemma-4-12B-it-OnlyText-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 OnlyTextLLMs/gemma-4-12B-it-OnlyText-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 OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF with OpenClaw:

Start the llama.cpp server

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

gemma-4-12B-it-OnlyText-GGUF

GGUF quants of OnlyTextLLMs/gemma-4-12B-it-OnlyText — the text-only derivative of google/gemma-4-12B-it. The file is the text backbone alone: no draft head is embedded, so there is no in-file speculative decoding — plain decoding is what the numbers below measure.

Quant files

File	Size
`gemma-4-12B-it-OnlyText-Q4_K_M.gguf`	7.4 GB
`gemma-4-12B-it-OnlyText-Q6_K.gguf`	9.8 GB
`gemma-4-12B-it-OnlyText-Q8_0.gguf`	12.7 GB

Quantization damage (vs the F16 master)

llama-perplexity --kl-divergence against the F16 master's logits (ctx 1024, 40 windows, 20440 scored tokens). Same top-1 is the share of positions where the quant keeps the F16 file's top token; KLD is the divergence of the full distribution, whose median is the tail-free summary (a mean over a heavy-tailed loss says more about the tail than about the quant).

Quant	Mean KLD	Median KLD	Same top-1	Max KLD
`Q4_K_M`	1.4721	0.7812	56.27%	29.33
`Q6_K`	0.4131	0.0865	77.93%	22.61
`Q8_0`	0.2070	0.0224	85.80%	24.44

Q4_K_M is not interchangeable with the others here. It changes the top-1 token on 43.7% of positions, against 14.2% for Q8_0, so choose from the Same top-1 column rather than by size alone.

Benchmarks

Measured 2026-09-24 on a single AMD Radeon AI PRO R9700 (gfx1201, 34 GB) with llama.cpp 9d286e1b315c (HIP/ROCm build, ROCm 7.14, -c 32768 -fa on). Baseline = llama-bench PP512/1024/2048 + TG128/256; chat decode = llama-cli single-turn, 256 generated tokens, median conditions, -np 1.

Quant	Device	Prefill t/s (PP512)	Gen t/s chat	Gen t/s llama-bench TG128
Q4_K_M	single R9700	2030	54.4	57.9
Q6_K	single R9700	1347	46.2	47.6
Q8_0	single R9700	1798	38.5	39.5

The two decode columns are different measurements (chat runs use the chat template and a ~90-token prompt; TG128 is a bare 128-token generation) — do not compare them against each other.

No published baselines are cited for this size — the numbers above are ours.

Usage

Requires a llama.cpp build that knows the gemma4 architecture.

# llama-server (OpenAI-compatible)
llama-server -m gemma-4-12B-it-OnlyText-Q4_K_M.gguf -ngl 99 -c 32768 -fa on

# llama-cli single-turn chat
llama-cli -m gemma-4-12B-it-OnlyText-Q4_K_M.gguf -ngl 99 -c 32768 -fa on \
    -st -p "Your prompt"

Notes:

The chat template supports Gemma's thinking channel but defaults enable_thinking to false, so plain chat answers directly. Pass "chat_template_kwargs": {"enable_thinking": true} (server API) to let it reason first.
Dual-GPU: unset *_VISIBLE_DEVICES env pins; llama.cpp auto layer-splits (--split-mode layer is the default).

Details

Base: google/gemma-4-12B-it (apache-2.0): 11.91B params, 48 layers (40 sliding-attention + 8 full-attention), hidden 3840, vocab 262137, Gemma4UnifiedForCausalLM, gemma4 GGUF architecture.
Tokenizer: 17 text-only special tokens at ids 0–106 (EOS 1 = <eos>, turn end 106 = <turn|>); image/audio/video tokens removed.
Conversion: convert_hf_to_gguf.py (llama.cpp 9d286e1b315c) then llama-quantize with the file's named type; no draft head is embedded (the family ships none).
Weights are untouched derivatives — no additional training.

Attribution

This model is a derivative of google/gemma-4-12B-it by the Google team, released under the apache-2.0 license. All credit for the underlying weights and capabilities belongs to the original authors; this repository only removes modalities and quantizes, it does not add new training.

Accuracy against the source checkpoint

Scored per position — this file's logits against the source checkpoint at matching precision, over the same 1533 tokens (ctx 1024, 3 windows of wikitext-2) — the two agree on the top-1 token at 90.6% of positions, median KLD 0.005. Both sides hold the same weights in the same precision there, so that gap is llama.cpp against transformers, not quantization: on this family's out-of-distribution corpus (untemplated prose, instruct-only model) the top-1 token sits near a tie at many positions, and any numerical difference — kernel, dtype, device — flips it, while the median says the two distributions are otherwise nearly identical.

Quantization costs a further 4.2 points of top-1 agreement (Q8_0 86.4%, measured the same way), which is the axis the table above is for.

Downloads last month: 39

GGUF

Model size

12B params

Architecture

gemma4

Hardware compatibility

4-bit

6-bit

8-bit

Model tree for OnlyTextLLMs/gemma-4-12B-it-OnlyText-GGUF

Base model

google/gemma-4-12B

Finetuned

google/gemma-4-12B-it

Quantized

(323)

this model