Inkling 2 Mini, Cutting-Edge Performance for Your Local Setup (huggingface.co)

🤖 AI Summary
Thinking Machines Lab has unveiled the Inkling 2 Mini, an innovative multimodal model that employs a looped transformer architecture. This model features approximately 26 billion total physical parameters but activates around 3 billion per token through a unique mechanism that allows certain layers to be revisited three times, effectively enhancing its performance without increasing the physical parameter count. With a substantial context size of 1 million tokens, Inkling 2 Mini accepts both text and image inputs and includes mixed-precision formats for efficient performance, establishing it as a solid choice for local AI setups. This release holds great significance for the AI/ML community as it represents a leap in multimodal processing capabilities while maintaining a smaller footprint than its predecessor, Inkling 2 Small. It has already shown competitive results across numerous benchmarks, outperforming other models like Qwen 3.8 and Gemma 4 in areas such as coding, reasoning, and vision interpretation. The Inkling 2 Mini also offers practical integrations through frameworks like transformers and Llama App, making it accessible for developers to leverage advanced features in local environments. Overall, Inkling 2 Mini is positioned as a powerful and resource-efficient tool for a range of AI applications.
Loading comments...
loading comments...