Is kscale-kinfer suitable for real-time applications?

Absolutely. It is specifically designed for real-time use cases, featuring a single Tokio runtime and input pre-fetching to minimize execution jitter and latency.

Does this skill support Python development?

Yes, while the core engine is written in Rust for performance, it includes Python bindings via PyO3, allowing you to use the high-performance engine within Python-based workflows.

What model formats can I use with this skill?

The skill is built around the ONNX format, which allows you to deploy models exported from popular frameworks like PyTorch, TensorFlow, or JAX.

How does kscale-kinfer handle telemetry?

It utilizes an asynchronous logging thread that records data in NDJSON format, ensuring that debugging and telemetry collection do not interfere with the timing of the inference loop.

What is kscale-kinfer used for?

It is a specialized engine used to run trained machine learning models, specifically ONNX policies, on robotic hardware with the low latency required for real-time control.

K-Scale Kinfer Inference

Name: K-Scale Kinfer Inference
Author: plurigrid

byplurigrid

•

数据科学与机器学习

Deploys high-performance reinforcement learning policies to real-world robots using a specialized Rust-based ONNX inference engine.

K-Scale Kinfer provides a specialized capability for Claude to assist with the deployment and optimization of neural network policies on robotic hardware. By leveraging a high-performance Rust implementation and the ONNX runtime, it enables the low-latency inference and real-time execution necessary for physical motion control. This skill guides users through loading ONNX models, managing asynchronous telemetry via NDJSON, and integrating inference loops with KOS firmware, effectively bridging the gap between trained machine learning models and embedded robotic systems.

主要功能

01Comprehensive telemetry and debugging via NDJSON logging

02Real-time policy execution with asynchronous input pre-fetching

03High-performance Python bindings via PyO3 for hybrid development

04Seamless integration with K-Scale's KOS firmware and robotic systems

058 GitHub stars

06Low-latency ONNX model inference using Rust and the ort runtime

使用场景

01Deploying trained walking or manipulation policies to physical humanoid robots

02Optimizing neural network inference loops for low-latency embedded systems

03Debugging real-time robot sensor data and action outputs using NDJSON telemetry

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add plurigrid/asi kscale-kinfer

For use in Claude.ai and ChatGPT

Download Skill