LLM Reference

RecurrentGemma 2B

About

RecurrentGemma 2B, developed by Google, leverages a novel Griffin architecture that integrates linear recurrences with local attention mechanisms to adeptly manage long sequences while minimizing memory usage. It excels in text generation tasks, such as question answering, summarization, and reasoning, by efficiently handling complex prompts and instructions. Available in both pre-trained and instruction-tuned versions, RecurrentGemma enhances usability in interactive applications like chatbots. Its open-source nature fosters transparency, enabling researchers to explore and innovate further. Performance-wise, it stands out with competitive results on benchmarks like HellaSwag and PIQA, marking a notable leap in natural language processing capabilities.

Capabilities

MultimodalFunction CallingTool UseJSON Mode

Providers(1)

ProviderInput (per 1M)Output (per 1M)Type
NVIDIA NIMProvisioned

Specifications

Released2024-04-09
Parameters2B
ArchitectureDecoder Only
Specializationgeneral