Intellectual hub of the topic
The Pruna AI team has accelerated image generation in the FLUX.2 [flex] model threefold without compromising quality. We explain how this was achieved and what it means for users.
OpenHands has launched a benchmark demonstrating how models handle real-world GitHub tasks – from bug fixes to implementing new features in open-source projects.
AI: Events
Technical context • Infrastructure
Engineers at AI21 Labs discovered a bizarre bug in vLLM that turned the Jamba model's normal responses into gibberish – and it was all down to a single incorrect token.
YouTube creators can now leverage AI avatar technology to produce short videos, thanks to a new platform tool powered by Supertone.
AI: Events
Development
AI21 Labs demonstrated that a single «chunk» size in RAG systems is a compromise and proposed a simple way to adapt text segmentation to the user's query type.
AI: Events
Technical context • Development
Anthropic has enhanced Claude's capabilities in handling low-level code and transferring knowledge to other models through its new «Extended Thinking» feature.
AMD has introduced a tool for automatically identifying the best quantization settings for ONNX models, eliminating the need for developers to manually sift through options.
AMD has demonstrated how to deploy OpenHands – an agent for automating code writing – on its server GPUs using the vLLM engine.
Cursor found a way to speed up the indexing of large codebases by safely reusing indexes created by colleagues, reducing the time from hours to seconds.
Want to know about new
experiments first?
Subscribe to our Telegram channel — we share all the latest
and exciting updates from NeuraBooks.