gemma-4-E4B-it-F32-GGUF

gemma-4-E4B-it-F32-GGUF

Gemma-4-E4B-it from Google is a 4.5B effective parameter (8B total with Per-Layer Embeddings) multimodal dense model in the Gemma 4 family, optimized for edge deployment on laptops, high-end smartphones, and consumer GPUs with native support for text, images (variable aspect ratio/resolution), audio processing, and configurable thinking modes for step-by-step reasoning. Featuring 42 layers, 512-token sliding window, 128K context length, and 262K vocabulary, it delivers frontier-level performance in agentic workflows, multilingual OCR/handwriting recognition, document/PDF parsing, UI/screen analysis, chart interpretation, object detection with pointing, coding assistance, and low-latency speech-to-text understanding—rivaling models 10-20x larger while maintaining Google's production-grade safety alignments. The instruction-tuned variant excels at on-device autonomous agents via Android AICore/Qualcomm optimizations, with open weights enabling local-first inference (MediaTek/ARM CPUs, NVIDIA RTX) for privacy-focused applications like mobile IDEs, real-time document processing, and structured data extraction in resource-constrained environments.

Quick start with llama.cpp

llama-server -hf prithivMLmods/gemma-4-E4B-it-F32-GGUF:F32

Model Files

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):