Hugging Face re-evaluated all 3,751 models on the Open LLM Leaderboard using the Math-Verify tool to fix math evaluation issues, resulting in an average 4.66-point score increase and significant…
Fireworks.ai is now a supported Inference Provider on the Hugging Face Hub, enabling serverless inference on models including DeepSeek-R1, DeepSeek-V3, Mistral-Small-24B-Instruct-2501,…
This blog breaks down how to calculate cost and latency for large scale classification and embedding tasks, analyzing different model architectures and hardware choices.
Hugging Face released open source tooling and scripts for building video generation datasets, featuring a three-stage pipeline for acquisition, filtering, and processing.
Hugging Face's Xet team announces a new chunk-based approach using blocks, shards, and key chunks to accelerate uploads and downloads on the Hub, with open-sourced components xet-core and hf_xet…
This blog post provides an update on the Open R1 project, introducing the OpenR1-Math-220k dataset generated on 512 H100s with DeepSeek R1, detailing the generation and filtering processes, and…
LG AI Research announced the successful conclusion of its 2024 courses at the LG AI Graduate School, an in-house program that trains AI experts through real-world projects.
LG AI Research published an AI ethics accountability report and participated in the AI Action Summit, highlighting its open-source models EXAONE 3.0 and 3.5, educational programs, and collaboration…
LG AI Research published Episode 2 of its AI Accountability Report, detailing governance, research, and engagement initiatives toward Responsible AI, including an AI Ethical Impact Assessment applied…