---
license: gemma
base_model: google/gemma-3-12b-it
pipeline_tag: translation
library_name: transformers
language:
- cs
- de
- en
- zh
- ar
tags:
- wmt26
- model-compression
- machine-translation
- gemma
- compressed-tensors
- vllm
---

# gemma-3-12b-wmt26-fp8

A **compressed, modified derivative** of [`google/gemma-3-12b-it`](https://huggingface.co/google/gemma-3-12b-it) for the
**WMT26 Model Compression** shared task (constrained track). This file has been modified from
the original Gemma 3 release: **FP8 dynamic (compressed-tensors)**.

- **Base model:** `google/gemma-3-12b-it` (Google)
- **Compression:** FP8 dynamic (compressed-tensors)
- **Translation directions:** ces-deu eng-zho_Hans eng-ara_EG
- **Serving:** vLLM, validated stack `vllm==0.23.0` / `transformers==5.12.1` /
  `compressed-tensors==0.17.0`. Inference entrypoint, environment, and run instructions are in
  the submission package (`run.sh` / `setup.sh` / `requirements.txt`).

## Built with Gemma — license & use

Gemma is provided under and subject to the **Gemma Terms of Use**, found at
https://ai.google.dev/gemma/terms . Use is additionally governed by the Gemma
**Prohibited Use Policy** (https://ai.google.dev/gemma/prohibited_use_policy). This is a
derivative of `google/gemma-3-12b-it`; Gemma and its trademarks are the property of Google. By using these
weights you agree to the Gemma Terms of Use.