Gemma4 E2B Int4.mog

Gemma4 E2B Int4.mog is a text generation model distributed in the .mog format and made available on Hugging Face. The model is based on the Gemma 4 E2B architecture, specifically an instruction-tuned variant that has undergone quantization-aware training (QAT) and is quantized to int4 precision. The provided package contains a single file that includes the configuration, tokenizer, and text weights, with a total size of approximately 2.5 GB. The model omits vision and audio capabilities, focusing exclusively on text processing. The quantization method uses group-wise symmetric int4 quantization (with a group size of 64) applied to attention and MLP projections as well as embedding tables, while normalization layers are retained in f16 format. The architecture features hybrid sliding and full attention mechanisms, dual rotary positional embeddings (RoPE), and key-value (KV) sharing. The context configuration allows for a maximum of 131,072 position embeddings, indicating support for long text sequences. Gemma4 E2B Int4.mog is distributed under the Apache 2.0 license, following the licensing of the underlying Gemma 4 base model. The evidence does not specify intended user roles or particular deployment platforms, but it does mention that the model is text-only and provides instructions for exporting the model using Python scripts and requirements. No pricing information is given, and the model is positioned as an open-source resource within the class of text generation models.