Inkling Unveils Open-Weights Model with 975 Billion Parameters for LLM
fine-tuning multimodal reasoning
| Source: HN | Original article
Researchers unveil Inkling, a large multimodal AI model. Inkling is open-weights and features a Mixture-of-Experts architecture.
Thinking Machines Lab has unveiled Inkling, a groundbreaking 975B parameter Large Language Model (LLM) with open-weights, allowing for public access and customization. This Mixture-of-Experts transformer boasts 41B active parameters and supports a context window of up to 1M tokens, enabling controllable reasoning effort.
What makes Inkling significant is its open-weights nature, permitting users to fine-tune the model on their own and download it for self-hosting. This development matters because it democratizes access to advanced AI technology, potentially accelerating innovation and research in the field.
As the AI community begins to explore Inkling's capabilities, it will be interesting to watch how this open-weights model performs in various applications and benchmarks, particularly in comparison to other LLMs. The release of a smaller companion model, Inkling-Small, with 276B parameters, also presents an opportunity to evaluate the trade-offs between model size and performance.
Sources
Back to AIPULSEN