This is a highly optimized, extremely compressed 4-bit (IQ4XS) GGUF version of Google’s Gemma 4 31B Instruct. The quantization was performed using a custom Importance Matrix (imatrix)** calculated over 2200+ chunks. This ensures maximum retention of the model’s analytical, mathematical, and reasoning capabilities, while aggressively shrinking the file size from ~61 GB (FP16) down to just ~15.58 GB. Thanks to the preserved 262,144 token vocabulary and custom imatrix, this model exhibits exceptional multilingual logic. Supported languages include (but are strictly not limited to): English, Russian, German, French, Italian, Mandarin Chinese, Japanese, Korean, and Arabic. During testing, the model successfully performed highly technical translations involving complex alphabets and provided unprompted transliteration (Pinyin, Romaji) and literal translation breakdowns. Base Model: Google Gemma 4 31B Instruct Quantization Format: IQ4XS (GGUF) Optimization: Custom Imatrix applied File Size: ~15.58 GB Tested Context Size: 250,000 tokens (using Flash Attention) Vocabulary Size: 262,144 tokens License: Gemma This model is fully compatible with llama.cpp and frontends like Open WebUI.…
Модальности:
Генерация текста
Области применения:
Диалог / чат RAG (поиск + генерация)
Задача: Генерация текста
Автор: Krasnopjorovs
Теги: gguf, gemma, gemma-4, imatrix, quantization, rag, llama-cpp, en
Лайков: 3 | Загрузок: 106
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
