The listed file size is 5.15 GB. Google says QAT is intended to reduce memory requirements while preserving quality similar to bfloat16. The card describes E4B as accepting text, image, and audio input, with a 128K-token context window. This is a ready-to-deploy GGUF distribution; the card does not establish device-specific performance or independent quality results for this checkpoint.
Raw SignalOct 14, 20261 min read1 source
Google released Gemma 4 E4B QAT Q4_0 GGUF , a 5.15 GB on-device multimodal model.
In one minute
Google’s Hugging Face model card lists an instruction-tuned Gemma 4 E4B checkpoint using Quantization-Aware Training (QAT) in GGUF Q4_0 format.