Gemma: Introducing new state-of-the-art open models

Abstraction: Google DeepMind's Gemma open-weight model family launch announcement

Key points:

• Gemma is a family of lightweight open models (2B and 7B) built from the same research and technology as Gemini; both pre-trained and instruction-tuned variants released.
• Surpasses significantly larger open models on key benchmarks while running on a developer laptop or desktop.
• Safety measures include automated filtering of personal/sensitive data from training sets, RLHF alignment, red-teaming, and adversarial testing.
• Supports JAX, PyTorch, TensorFlow/Keras 3.0; integrates with Hugging Face, NVIDIA NeMo, TensorRT-LLM; deployable on Vertex AI and GKE.
• Commercial use permitted for all organization sizes; free access via Kaggle and Colab; up to $500,000 in Google Cloud credits available for researchers.

Connections: Gemma · Google · Google Deepmind · Open Source Models · Large Language Models · Responsible AI

Source: https://blog.google/technology/developers/gemma-open-models/