Thinking Machines Launches Inkling 2 Small MoE Model
Thinking Machines Lab has released Inkling 2 Small, a 594B MoE model with 17B active parameters, running on Tinker with Apache 2.0 weights planned.
Stock photo for illustration only, not from the actual event
- Thinking Machines Lab released the Inkling 2 Small AI model
- Features a 594B parameter MoE architecture with 17B active parameters
- Achieves a score of 31 on the Artificial Analysis Intelligence Index
- Operates on Tinker with upcoming Apache 2.0 weight availability
Artificial intelligence research group Thinking Machines Lab has officially introduced Inkling 2 Small, a new model built on a Mixture of Experts (MoE) architecture designed to enhance processing efficiency and computational performance for advanced workloads.
The newly unveiled model features a total parameter count of 594 billion (594B), with 17 billion active parameters (17B) engaged during runtime. This design allows the system to balance a vast knowledge base with optimized resource allocation during inference tasks.
Stock photo for illustration only, not from the actual event
Regarding benchmark evaluations, Inkling 2 Small scored 31 on the Artificial Analysis Intelligence Index, demonstrating its competency and performance metrics under standardized artificial intelligence assessments.
Mixture of Experts (MoE) models continue to represent a crucial strategy in modern AI development. By routing tokens to specific expert sub-networks rather than activating the entire network for every calculation, MoE models achieve high capacity and broad knowledge storage without incurring the massive computational cost of dense models of equivalent total size.
The model operates natively on Tinker, and the laboratory has announced plans to release the model weights under an Apache 2.0 license, which will facilitate open access for developers and researchers looking to integrate or build upon the system.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment