Hugging Face

Gradually adding evaluations for models in Transformers library.

Allows easy sharing of leaderboard images on social media.

Supports new AI methods based on popularity and citations.

Displays follow-up or predecessor papers with a banner.

Submit papers from various sources beyond Arxiv for enrichment.

Leaderboards now support multiple metrics for AI benchmarks.

New Sentence Transformers CrossEncoder rerankers for improved text relevance.

Framework for running and reproducing AI agent evaluations.

An open benchmark for comparing full AI agent systems.

Enhances GPU utilization by enabling parallel CPU-GPU operations.

Full-size model with Matryoshka dimension support and 32K context.

Compact model with high retrieval quality and 32K-token context.

Substantial rewrite of vLLM engine for RL correctness.

FP8 quantized variants optimized for inference with vLLM.

8B instruct model surpasses previous Granite 4.0-H-Small.

Dense, decoder-only LLMs trained on 15 trillion tokens.

Runs systematic tests on converted code for reproducibility.

Aids in porting language models from Transformers to MLX.

Hugging Face supports small OCR models for low-end devices.

Maps inputs from different modalities into shared embedding space.

Score relevance between multimodal input pairs using AI.

Models map text, images, audio into shared embedding space.

Mutable storage solution for AI data without versioning issues.

Serverless compute platform supporting CPUs and GPUs for AI.

Enables interaction with papers using HuggingChat AI technology.