NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·3h ago
Data Science
Analytics Vidhya·16d ago·5 min read
Complete Guide to Thinking Machines Inkling
TL;DR
Thinking Machines Lab has launched Inkling, a general-purpose foundation model with 975 billion parameters designed for multimodal reasoning and various AI tasks. Unlike models focused solely on benchmarks, Inkling emphasizes customization for specific applications, including coding and tool use.
✦ Why It Matters
Engineers can start integrating Inkling into their projects for customized AI solutions today.
Key Takeaways
- Utilize Inkling's multimodal capabilities for diverse AI tasks.
- Leverage the open-weights model for custom fine-tuning.
- Implement Inkling in applications requiring text, image, and audio processing.
- Explore integration options with existing tools and frameworks.
How It Works
Inkling employs a Sparse Mixture-of-Experts architecture, activating only a fraction of its parameters per token to reduce computational costs. Each token's processing involves routing through 6 experts, optimizing performance while maintaining flexibility across various input modalities.