TranslateGemma 27B, int8 -- radiance container This repository holds a Model Derivative of Google's TranslateGemma 27B (https://huggingface.co/google/translategemma-27b-it). Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. Use of this file is subject to those terms, including the Gemma Prohibited Use Policy (ai.google.dev/gemma/prohibited_use_policy), which is incorporated in them. Anyone who passes this file on must pass these terms on with it. Changes from the original weights: every attention and feed-forward linear and the output head were quantised to int8 with a bf16 scale per 128 columns (round to nearest); the vision tower and the multimodal projector were left out; the tensors were written into radiance's .rad container format. No weight was trained or otherwise altered.