Gemma4-12B-it-Fast
A community quantized derivative published by DevelopingDad.

Original Gemma 4 model, architecture, instruction-tuned model, tokenizer,
processor and native template: Google DeepMind / Gemma Team.
https://huggingface.co/google/gemma-4-12B-it

Immediate source: DevelopingDad/gemma-4-12b-it-trial62
Revision: 568aa9764b4f695fb18881b97f6d4ee8b725f75c
This export and release integration: DevelopingDad, 2026.

Optional native assistant: Google DeepMind / Gemma Team.
https://huggingface.co/google/gemma-4-12B-it-assistant
Its weights are obtained separately.

Optional runtime code includes modified vLLM project files.
Copyright and SPDX notices in those files are retained.
Upstream: https://github.com/vllm-project/vllm
Revision: db9527a46873454610df6dbedf79a36d6bf1a7f6
vLLM and the model are licensed under Apache License 2.0; see LICENSE.

Quantization tooling: llm-compressor and compressed-tensors contributors
(vLLM Project / Red Hat AI).
Calibration corpus: Google C4, processed and hosted by AllenAI, based on
Common Crawl. C4 is ODC-BY licensed; corpus text is not redistributed.
Runtime/kernel/framework credit: vLLM, FlashInfer, NVIDIA CUTLASS/CUDA,
PyTorch and Hugging Face Transformers contributors.
Prior serving comparisons: Unsloth and Red Hat AI NVFP4 checkpoints.

This release is independent of and not endorsed by those upstream projects.
These notices are informational and do not modify the applicable licenses.