This is an NVFP4-quantized GGUF version of google/gemma-4-26B-A4B-it, Google’s state-of-the-art 26B-parameter MoE instruction-tuned model (4B active parameters, 128 experts, 8 experts per token). NVFP4 is NVIDIA’s native Blackwell FP4 quantization format, delivering excellent quality at 4-bit precision with optimized hardware support on Blackwell GPUs. This repository includes the multimodal vision encoder (mmproj) for image understanding capabilities. — Thinking is disabled by default. To enable thinking, pass enablethinking=true` in your request. — Gemma-4 uses a hybrid attention mechanism with sliding window (local) and global (full) attention layers in a 5:1 pattern. — The model uses proportional RoPE for global attention layers and standard RoPE for sliding window layers. — NVFP4 requires llama.cpp built with CUDA support and a Blackwell-compatible GPU for best performance. — CPU inference is supported but slower than GPU-accelerated inference. — GPU: Blackwell (RTX 5090, B200, etc.) for optimal NVFP4 performance — RAM: 8 GB+ (model fits in most modern GPUs at 4-bit) — Storage: ~14 GB for the model files — Google for the original Gemma 4 model — llama.cpp for the GGUF format…
Модальности:
Генерация текста Компьютерное зрение Мультимодальность
Области применения:
Следование инструкциям Диалог / чат
Задача: Генерация текста
Автор: FreedomAISVR
Теги: gguf, google, gemma-4, nvfp4, 4-bit, blackwell, fp4, moe
Лайков: 3 | Загрузок: 337
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
