We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hacker News·20d ago·5 min read
Inkling: Our Open-Weights Model
TL;DR
Inkling is a new open-weights AI model designed for customization, featuring 975 billion parameters and multimodal capabilities. It allows users to fine-tune the model for specific applications using the Tinker platform.
✦ Why It Matters
Engineers can start fine-tuning Inkling today using the Tinker platform to create tailored AI solutions.
Key Takeaways
- Explore Inkling's multimodal capabilities for diverse applications.
- Utilize the Tinker platform for easy model fine-tuning.
- Leverage controllable thinking effort to optimize performance.
- Experiment with Inkling-Small for cost-effective solutions.
Full Summary
Inkling is a Mixture-of-Experts transformer model with 975 billion total parameters, of which 41 billion are active, and it supports a context window of up to 1 million tokens. Trained on 45 trillion tokens of diverse data, it excels in multimodal reasoning across text, images, and audio.
Inkling is designed to be a flexible foundation model, allowing users to customize it for specific tasks using the Tinker platform. A lighter version, Inkling-Small, with 12 billion active parameters, is also available for lower-cost applications.
The model demonstrates strong performance in agentic coding and tool use, achieving competitive scores on various benchmarks. Its controllable thinking effort feature allows developers to balance performance and efficiency, making it suitable for real-world applications.