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

<!-- SC_OFF --><div class="md"><p>Google DeepMind dropped Gemma 4 today:</p> <p><strong>Gemma 4 31B:</strong> dense, 256K context, redesigned architecture targeting efficiency and

<p>Today Inception Labs dropped <strong>Mercury 3</strong>: a diffusion-based LLM with a 1M-token context window doing <strong>1,068 tokens/sec on NVIDIA GPUs</strong>, claiming

<a href="https://news.ycombinator.com/item?id=48478471">Comments</a>

<a href="https://news.ycombinator.com/item?id=48321076">Comments</a>

<!-- SC_OFF --><div class="md"><p><a href="https://huggingface.co/google/gemma-4-31B-it-assistant">https://huggingface.co/google/gemma-4-31B-it-assistant</a></p> <p><a

Gemma 4: Our most intelligent open models to date, purpose-built for advanced reasoning and agentic workflows.

<!-- SC_OFF --><div class="md"><p><a href="https://huggingface.co/unsloth/gemma-4-E2B-it-GGUF">https://huggingface.co/unsloth/gemma-4-E2B-it-GGUF</a></p> <p><a

<!-- SC_OFF --><div class="md"><p>The secret sauce here is that the student model does not just try to guess the next token in a sentence, which is how most AI is trained.