Brief

GitLab has introduced "Orbit," a knowledge graph that structures code repositories to provide agents with precise context, addressing the issue of high token costs associated with large language models. This approach, validated by GitLab's own development team, significantly reduces the amount of data agents need to process, thereby cutting costs. The article contrasts GitLab's solution with a local, MCP-native alternative called jCodeMunch, which offers similar efficiency benefits without requiring enterprise-level infrastructure or recurring fees. AI

Sphere AI Lab has released Orbit, an open-source framework designed for the post-training of large language models. This framework allows for the fine-tuning of trillion-parameter models, such as DeepSeek-V4, to be conducted on a single high-performance node. The development aims to make advanced model training more accessible. AI

Orbit has released an NPM SDK designed to streamline AI query routing for developers. This tool automatically directs requests to the most suitable AI model, eliminating the need for manual selection. The SDK aims to simplify the process of leveraging multiple AI models within applications. AI

Researchers have introduced ORBIT, a new method designed to prevent large language models from losing their foundational language capabilities during task-specific fine-tuning. This issue, known as catastrophic forgetting, is particularly prevalent in Generative Retrieval tasks and is linked to the divergence of model parameters. ORBIT addresses this by monitoring the distance between fine-tuned and original model weights, employing a weight averaging strategy to limit parameter drift when a set threshold is exceeded. Experiments demonstrate that ORBIT effectively preserves text and retrieval performance, outperforming existing continual learning and regularization techniques. AI