Inkling 2 Small favicon

Inkling 2 Small

Inkling 2 Small is a 594B parameter mixture-of-experts language model with 17B active parameters and a 256K context window.

TextProcessing and generating text across…Serving via serverless inference with…Training and sampling via Tinker's…Mixture-of-Experts Architecture
Inkling 2 Small product interface screenshot
Estimated monthly visits
376K
Data period:
Listed on AIToolly

What Is Inkling 2 Small? Product Overview

What the product does and how it is positioned

Inkling 2 Small is a frontier-scale foundation model developed by Thinking Machines. Built on a mixture-of-experts architecture, the model pairs 594 billion total parameters with 17 billion active parameters, offering an extended maximum context window of 256K tokens.

Hosted on the Tinker platform, Inkling 2 Small is available for training, sampling, and serverless inference in beta. It can be integrated into custom workflows through Tinker's API and client SDK.

What Can You Use Inkling 2 Small For?

Source-supported ways to use the product

Custom Post-Training and Fine-Tuning

Organizations can adapt Inkling 2 Small on specialized domain datasets using supervised fine-tuning or reinforcement learning through the Tinker training API.

Long-Context Serverless Text Inference

Developers can deploy Inkling 2 Small through serverless inference endpoints to process lengthy documents up to 256K tokens with cached inputs.

Model Architecture and Platform Integration

Inkling 2 Small implements a sparse mixture-of-experts architecture consisting of 594B total parameters, routing computation so that only 17B parameters remain active per token. This design provides broad representational capacity while containing active inference and training compute requirements.

The model integrates directly into Thinking Machines' Tinker platform, which coordinates gradient accumulation, forward-backward passes, optimizer steps, and periodic checkpoint management across managed GPU clusters.

  • Mixture-of-experts architecture with 594B total parameters and 17B active parameters.
  • Context length capacity supporting up to 256K tokens.
  • Available for serverless inference during the beta phase with input caching.
  • Compatible with Tinker's managed training loop, periodic checkpointing, and weight export endpoints.

What to Test Before Choosing Inkling 2 Small

Checks to run with your own material and workflow

  • Confirm whether the 256K token context window accommodates your organization's document sizes.
  • Verify that the 17B active parameter computational profile meets your operational latency requirements.
  • Check that serverless inference beta features and SLA levels meet your production deployment criteria.

Inkling 2 Small Sources and Last Checked

What was checked and when

Last checked
Category
Text

Inkling 2 Small Frequently Asked Questions

Answers based on the source-checked product record

What is the architecture of Inkling 2 Small?

Inkling 2 Small is a mixture-of-experts model with 594 billion total parameters and 17 billion active parameters.

What context length does Inkling 2 Small support?

Inkling 2 Small supports a maximum context window of up to 256K tokens.

Can Inkling 2 Small be accessed via serverless inference?

Yes, Inkling 2 Small is supported under Tinker's serverless inference offering, which is currently in beta.

Can saved model checkpoints be downloaded?

Yes, Tinker provides an API endpoint that allows users to download any checkpoint they have saved.

What security mechanisms protect workloads on Tinker?

Workloads are protected via customer-managed Cloud KMS encryption, VPC PrivateLink, custom OIDC issuers, role-based access controls, and organization audit logs.

Explore other recently added tools in the same category.

View all Text tools