turintech/ gemma-4-12b-it-artemis:q4_0

140 Downloads Updated 5 days ago

Text-only Gemma 4 12B GGUF files. Default: q4_0. TurinTech models, built with Artemis. Measured Q4_0: Mean KLD 0.156, Same top p 86.8% vs BF16 on WikiText-2 test; protocol a100-2026-10-24.

tools thinking

ollama run turintech/gemma-4-12b-it-artemis:q4_0

curl http://localhost:11434/api/chat \
  -d '{
    "model": "turintech/gemma-4-12b-it-artemis:q4_0",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='turintech/gemma-4-12b-it-artemis:q4_0',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'turintech/gemma-4-12b-it-artemis:q4_0',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 5 days ago

5 days ago

573ebca24e9a · 6.7GB

model

archgemma4

·

parameters11.9B

·

quantizationQ4_0

6.7GB

license

Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR US

11kB

params

{ "stop": [ "<bos>", "<|turn>", "<turn|>", "<|turn>user" ],

133B

template

{{ if .System }}<bos><|turn>system {{ .System }}<turn|> {{ end }}{{ if .Prompt }}<|turn>user {{ .Pro

159B

Readme

Gemma 4 12B Artemis

Google’s Gemma 4 12B for chat, reasoning and tool use on one GPU, from Google’s quantization-aware (QAT) checkpoint.

6.72 GB download for the default file, gemma-4-12B-it-Artemis-Q4_0.gguf.
On chat text it picks the original’s next token 90.3% of the time, against 77.8% for Google’s own Q4_0 (6.98 GB).

Run with Ollama

Tags: latest, q4_0 and 6.7gb are the same file, gemma-4-12B-it-Artemis-Q4_0.gguf (6.72 GB download).

ollama run turintech/gemma-4-12b-it-artemis:q4_0

This package holds the main file only. For images, use Ollama 0.40.11 or newer and pull from Hugging Face, which brings the repository’s vision projector along:

ollama run hf.co/turintech/gemma-4-12B-it-Artemis-GGUF:Q4_0

Results

Our file against Unsloth’s file of the same size, Google’s own Q4_0 and the original, on an NVIDIA A100. The tool-call and maths items were kept out of development and used in a final check fixed in advance.

File	Size	Same next token, chat	Tool calls (of 200)	GSM8K (of 250)
Ours, Q4_0	6.72 GB	90.3%	190	240
Unsloth UD-Q4_K_XL	6.72 GB	89.8%	191	242
Google Q4_0	6.98 GB	77.8%	190	238
BF16 original (QAT)	23.83 GB	reference	189	242

No paired test separated these files on tool calls or maths. How it was tested.

Files

Every file we publish for this model. Sizes are download sizes, not memory.

File	Size	Use it for	Ollama tag
`gemma-4-12B-it-Artemis-Q4_0.gguf`	6.72 GB	Default	`q4_0` (latest)

The Ollama tag runs our package, which holds the same file: ollama run turintech/gemma-4-12b-it-artemis:<tag>.

The vision projector and the MTP drafter are optional. The drafter is there twice, in MTP/ and at the top level, where -hf finds it.

Limitations

The numbers come from llama.cpp, not from Ollama or LM Studio.

Licence

Apache-2.0 (licence). See the Hugging Face repository for LICENSE and NOTICE.

# Gemma 4 12B Artemis

Google's Gemma 4 12B for chat, reasoning and tool use on one GPU, from Google's quantization-aware (QAT) checkpoint.

- 6.72 GB download for the default file, `gemma-4-12B-it-Artemis-Q4_0.gguf`.
- On chat text it picks the original's next token 90.3% of the time, against 77.8% for Google's own Q4_0 (6.98 GB).

## Run with Ollama

Tags: `latest`, `q4_0` and `6.7gb` are the same file, `gemma-4-12B-it-Artemis-Q4_0.gguf` (6.72 GB download).

```bash
ollama run turintech/gemma-4-12b-it-artemis:q4_0
```

This package holds the main file only. For images, use Ollama 0.40.11 or newer and pull from Hugging Face,
which brings the repository's vision projector along:

```bash
ollama run hf.co/turintech/gemma-4-12B-it-Artemis-GGUF:Q4_0
```

## Results

Our file against Unsloth's file of the same size, Google's own Q4_0 and the original, on an NVIDIA A100. The tool-call and maths items were kept out of development and used in a final check fixed in advance.

| File | Size | Same next token, chat | Tool calls (of 200) | GSM8K (of 250) |
| --- | --- | --- | --- | --- |
| **Ours, Q4_0** | 6.72 GB | 90.3% | 190 | 240 |
| Unsloth UD-Q4_K_XL | 6.72 GB | 89.8% | 191 | 242 |
| Google Q4_0 | 6.98 GB | 77.8% | 190 | 238 |
| BF16 original (QAT) | 23.83 GB | reference | 189 | 242 |

No paired test separated these files on tool calls or maths. [How it was tested](https://huggingface.co/turintech/gemma-4-12B-it-Artemis-GGUF#how-it-was-tested).

## Files

Every file we publish for this model. Sizes are download sizes, not memory.

| File | Size | Use it for | Ollama tag |
| --- | --- | --- | --- |
| [`gemma-4-12B-it-Artemis-Q4_0.gguf`](https://huggingface.co/turintech/gemma-4-12B-it-Artemis-GGUF/resolve/main/gemma-4-12B-it-Artemis-Q4_0.gguf) | 6.72 GB | Default | `q4_0` (latest) |

- The Ollama tag runs our package, which holds the same file: `ollama run turintech/gemma-4-12b-it-artemis:<tag>`.

The vision projector and the MTP drafter are optional. The drafter is there twice, in `MTP/` and at the top level, where `-hf` finds it.

## Limitations

- The numbers come from llama.cpp, not from Ollama or LM Studio.

## Licence

Apache-2.0 ([licence](https://ai.google.dev/gemma/docs/gemma_4_license)). See the Hugging Face repository for `LICENSE` and `NOTICE`.

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)