PublishersGoogle

Google

4 documents read from this publisher.

Gemma 4 QAT models: Optimizing model compression for mobile and laptop efficiency

first party release · official blog · 2026-06-05 · 10 techniques

Quantization-Aware Training Post-training quantization Quantization Channel-wise quantization Embedding and KV-cache quantization Mobile-specialized quantization schema Modality-specific deployment Q4_0 Quantization Format Static activation quantization Targeted 2-bit quantization

Introducing Gemma 4 12B: a unified, encoder-free multimodal model

first party release · official blog · 2026-06-03 · 4 techniques

Multi-Token Prediction Encoder-free multimodal architecture Raw audio projection into the LLM embedding space Lightweight vision embedding module

Accelerating Gemma 4: faster inference with multi-token prediction drafters

first party release · official blog · 2026-05-05 · 5 techniques

Speculative decoding Multi-Token Prediction efficient clustering in the embedder Increasing batch size for inference KV-cache sharing between drafter and target

Gemma 4: Byte for byte, the most capable open models

first party release · official blog · 2026-04-02 · 5 techniques

Mixture of Experts Fine-tuning Quantization Dense model architecture Effective parameter activation for edge inference