PublishersGoogle DeepMind

Google DeepMind

2 documents read from this publisher.

google/gemma-4-26B-A4B-it · Hugging Face

first party release · model card · 2026-07-02 · 17 techniques

Mixture of Experts Hybrid Attention Sparse expert activation Encoder-free multimodal architecture Per-Layer Embeddings (PLE)Configurable visual token budget CSAM filtering Evaluation without safety filters Function calling Sensitive data filtering Standardized sampling configuration Content quality and safety filtering Continuous monitoring Human evaluation Interleaved multimodal input Native system prompt support Step-by-step thinking mode

google/gemma-4-31B-it · Hugging Face

first party release · model card · 2026-07-02 · 15 techniques

Mixture of Experts Hybrid Attention Sparse expert activation Encoder-free multimodal architecture Excluding prior thinking from conversation history Per-Layer Embeddings (PLE)Configurable visual token budget CSAM filtering Evaluation without safety filters Sensitive data filtering Standardized sampling configuration Control-token-enabled thinking mode Proportional Rotary Position Embedding Unified Keys and Values multimodal prompt content ordering