1351

OpenAI SWE-bench Verified release: human‑validated benchmark improves software‑engineering evaluation

OpenAI released SWE‑bench Verified, a human‑validated 500‑sample subset of the SWE‑bench software‑engineering benchmark that removes ambiguous issues and unfair tests, enabling more reliable evaluation of AI models’ coding abilities; GPT‑4o solves 33.2% of these samples, more than double its score on the original benchmark.

1352

Introduction to ggml

ggml is a lightweight, C/C++ machine learning library optimized for Transformer inference and on-device LLM execution across diverse hardware backends.

1353

Hugging Face Unified Tool Use API

Hugging Face has introduced a unified tool use API that allows developers to use the same code to implement tool calling across Mistral, Cohere, NousResearch, and Llama models.

1354

Falcon Mamba 7B Release Notes

The Technology Innovation Institute (TII) has released Falcon Mamba 7B, the first large-scale pure State Space Language Model (SSLM) that matches the performance of state-of-the-art transformer models while eliminating attention-based memory scaling issues.

1355

Qwen2-Audio Release Notes

Qwen2-Audio is an audio-language model capable of voice chat and audio analysis across more than eight languages and dialects, surpassing previous state-of-the-art performance on multiple benchmarks.

1356

Zico Kolter Joins OpenAI Board of Directors

OpenAI has appointed Zico Kolter, a professor and Director of the Machine Learning Department at Carnegie Mellon University, to its Board of Directors and Safety and Security Committee.

1357

Hugging Face acquires XetHub to upgrade Hub storage and collaboration

Hugging Face acquired XetHub to replace Git LFS with a more efficient storage backend, enabling incremental updates, trillion‑parameter model support, and better collaboration on massive AI datasets.

1358

GPT‑4o System Card – capabilities, safety mitigations, and risk assessment

OpenAI released the GPT‑4o System Card, detailing the omni‑model's capabilities, safety mitigations, and medium overall risk rating.

1359

Qwen2-Math Release Notes

Qwen has released Qwen2-Math, a series of specialized mathematical LLMs (1.5B, 7B, and 72B) that outperform several closed-source models, including GPT-4o, on math benchmarks.

1360

Rakuten and OpenAI Partnership: Leveraging Generative AI for Customer Insights

Rakuten is utilizing OpenAI's APIs, RAG, and Code Interpreter to transform unstructured data into automated customer service, review summaries, and B2B market insights.

1361

OpenAI Structured Outputs API feature announcement

OpenAI introduced Structured Outputs on August 6, 2024, a new API feature that guarantees model responses exactly match developer‑provided JSON Schemas, improving reliability for data‑centric applications.

1362

Hugging Face TextImage Augmentation pipeline release

Hugging Face and Albumentations AI released a TextImage Augmentation pipeline that jointly modifies document images and their text, enabling realistic synthetic data generation and robust fine‑tuning of vision‑language models on limited document datasets.

1363

Hugging Face 2024 Security Feature Highlights

Hugging Face has detailed its 2024 security landscape, introducing a suite of default protections for all users and advanced governance controls for Enterprise Hub users.

1364

Google releases Gemma 2 2B, ShieldGemma, and Gemma Scope

Google released Gemma 2 2B, ShieldGemma safety classifiers, and Gemma Scope sparse autoencoders, expanding open‑source LLM capabilities, moderation tools, and interpretability resources.

1365

Quanto quantization cuts memory for Transformer diffusion pipelines

Hugging Face quantization (Quanto) reduces GPU memory for Transformer diffusion models from ~12 GB to ~5 GB with minimal latency and quality impact.

1366

Hugging Face NVIDIA NIM API (serverless) launch and deprecation

Hugging Face launched the NVIDIA NIM API (serverless) for Enterprise Hub users, enabling pay‑as‑you‑go, serverless inference of open‑source LLMs on NVIDIA DGX Cloud H100 GPUs via an OpenAI‑compatible API.

1367

LAVE: Zero-shot VQA Evaluation on Docmatix with LLMs

Hugging Face introduces LAVE (LLM-Assisted VQA Evaluation) to address the rigidity of traditional VQA metrics, demonstrating a 50% accuracy gain in evaluating zero-shot performance on the Docmatix dataset.

1368

OpenAI SearchGPT Prototype Announcement

OpenAI has introduced SearchGPT, a temporary prototype that combines AI models with real-time web information to provide direct answers with clear source citations.

1369

OpenAI Rule-Based Rewards for Model Safety

OpenAI has introduced Rule-Based Rewards (RBRs), a method that uses explicit, step-by-step rules to align AI safety behavior without requiring extensive human data collection.

1370

Llama 3.1 Release Notes: Multilinguality, Long Context, and 405B Model

Meta has released Llama 3.1, featuring models in 8B, 70B, and 405B sizes with 128K context length, multilingual support for 8 languages, and a permissive license allowing synthetic data generation.

1371

Running Mistral 7B with Core ML

Hugging Face demonstrates how to run Mistral 7B on Mac using new Core ML features from WWDC 24, achieving a model size reduction to under 4GB using 4-bit block-wise quantization.

1372

GPT-4o mini release notes / what's new

OpenAI has released GPT-4o mini, a highly cost-efficient small model that outperforms GPT-3.5 Turbo and other small models on key reasoning and multimodal benchmarks.

1373

Docmatix Dataset Release

Hugging Face has released Docmatix, a Document Visual Question Answering (DocVQA) dataset featuring 2.4 million images and 9.5 million Q/A pairs, providing a 240x increase in scale over previous datasets.

1374

TGI Multi-LoRA: Deploy Once, Serve 30 Models

Hugging Face introduces Multi-LoRA serving in Text Generation Inference (TGI), allowing organizations to deploy a single base model and dynamically serve dozens of specialized fine-tuned adapters to reduce cost and operational complexity.

1375

ChatGPT Enterprise Compliance and Administrative Tools Update

OpenAI has introduced a Compliance API, SCIM for automated user management, and granular GPT controls to help enterprise customers meet regulatory requirements and scale AI deployments securely.

1376

OpenAI Prover‑Verifier Games improve legibility of language model outputs

OpenAI announced a prover‑verifier game training method that makes strong language models generate solutions that weaker models can verify, improving both correctness and human legibility.

1377

Argilla SDK Chatbot with distilabel – End‑to‑End Tutorial

Hugging Face released a tutorial showing how to build an Argilla 2.0 chatbot using distilabel‑generated synthetic data, fine‑tuned embeddings, lancedb vector storage, and a Gradio app deployed on Spaces.

1378

SmolLM Release: High-Performance Small Language Models

Hugging Face introduces SmolLM, a family of state-of-the-art small language models (135M, 360M, and 1.7B parameters) trained on a meticulously curated high-quality dataset.

1379

NuminaMath 7B TIR wins AIMO Progress Prize – technical recap

NuminaMath 7B TIR won the first AIMO Progress Prize by solving 29 of 50 hidden math problems, showcasing the power of a two-stage fine‑tuning recipe, large high‑quality math data, and a self‑consistency with tool‑integrated reasoning inference strategy.

1380

OpenAI and Los Alamos National Laboratory Bioscience Research Partnership

OpenAI and Los Alamos National Laboratory have partnered to evaluate how multimodal frontier AI models like GPT-4o can safely assist scientists in physical laboratory settings to advance bioscientific research.

1381

Hugging Face PII Detection Experiment with Presidio

Hugging Face is experimenting with integrating Microsoft Presidio into the Dataset Hub to provide automatic PII detection reports, helping practitioners identify and mitigate privacy risks in ML datasets.

1382

TRL adds Direct Preference Optimization support for Vision‑Language Models

Hugging Face added Direct Preference Optimization (DPO) support for Vision‑Language Models in the TRL library, enabling fine‑tuning of models like Idefics‑2 with preference data using bfloat16 quantization and LoRA to fit on a single GPU.

1383

Hugging Face and KerasHub Integration

Hugging Face and KerasHub now share a model save format, allowing KerasHub users to directly load over 300,000 Transformers library models from the Hugging Face Hub.

1384

Google Cloud TPUs on Hugging Face Inference Endpoints and Spaces

Hugging Face has integrated Google Cloud TPU v5e support into Inference Endpoints and Spaces, enabling users to deploy and scale AI models with cost-effective, high-performance hardware.

1385

Banque des Territoires, Polyconseil, and Hugging Face Deploy Sovereign RAG Solution for EduRénov Program

Hugging Face, Polyconseil, and Banque des Territoires launched a sovereign, open‑source RAG system to automate email support for France's EduRénov school‑renovation program, ensuring data residency while scaling generative AI for public policy.

1386

Hugging Face Dataset Hub Search Features Update

Hugging Face has introduced four new search filters—modality, size, format, and library compatibility—to improve the discoverability of over 180,000 public datasets on the Dataset Hub.

1387

Accelerating Protein Language Model ProtST on Intel Gaudi 2

Intel and MILA have optimized the ProtST multi-modal protein language model for Intel Gaudi 2 accelerators, achieving up to 2.92x faster fine-tuning compared to NVIDIA A100 GPUs.

1388

Hugging Face Transformers Code Agent GAIA Benchmark Results

Hugging Face achieved a top ranking on the GAIA benchmark using a Code Agent built with the transformers.agents library, demonstrating that code-based actions are more efficient and effective than JSON-based tool calling.

1389

Finding GPT-4’s mistakes with GPT-4

OpenAI introduced CriticGPT, a GPT-4 based model designed to identify errors in ChatGPT's code output to assist human trainers during the RLHF process.

1390

OpenAI and TIME Strategic Content Partnership

OpenAI and TIME have entered a multi-year strategic partnership to integrate TIME's 101-year archive of journalism into OpenAI products and provide TIME with access to OpenAI technology for product development.

1391

Google Gemma 2 Release Notes

Google has released Gemma 2, a family of open-weight LLMs available in 9B and 27B parameter sizes, featuring technical advances in distillation and attention mechanisms to improve performance.

1392

XLSCOUT ParaEmbed 2.0 Release

XLSCOUT has released ParaEmbed 2.0, a proprietary embedding model fine-tuned on expert-curated patent data that achieves a 23% increase in accuracy over ParaEmbed 1.0.

1393

Hugging Face Ethics and Society Newsletter #6: The Importance of Data Quality

Hugging Face outlines a holistic, responsible approach to data quality, emphasizing that high-quality data must be fit for its intended purpose to ensure AI model performance, fairness, and scientific reproducibility.

1394

Fine-tuning Microsoft Florence-2 for DocVQA

Hugging Face demonstrates how to fine-tune Microsoft's Florence-2 vision-language model on the DocVQA dataset, improving validation similarity from 0 to 57.0 after seven epochs.

1395

OpenAI Acquires Rockset for Real-Time Data Retrieval

OpenAI has acquired Rockset, a real-time analytics database, to integrate its high-performance indexing and querying capabilities into OpenAI's retrieval infrastructure.

1396

OpenAI Cybersecurity Grant Program: Supporting AI-Driven Defense Research

OpenAI has highlighted several projects from its Cybersecurity Grant Program aimed at enhancing LLM trustworthiness, automating vulnerability detection, and developing autonomous cyber defense agents.

1397

OpenAI Improved Techniques for Training Consistency Models

OpenAI has introduced improved consistency training techniques that allow generative models to learn directly from data without distillation, achieving significant FID score improvements on CIFAR-10 and ImageNet 64x64.

1398

OpenAI Consistency Models: Fast One-Step Generative AI

OpenAI has introduced Consistency Models, a new family of generative models that enable high-quality, one-step generation by directly mapping noise to data, overcoming the slow iterative sampling of diffusion models.

1399

OpenAI Holistic Approach to Undesired Content Detection

OpenAI has introduced a holistic framework for building robust natural language classification systems to detect undesired content, including sexual content, violence, and harassment, which outperforms off-the-shelf models.

1400

Hugging Face Data Is Better Together Initiative

Hugging Face and Argilla launched the Data Is Better Together (DIBT) initiative to empower the open-source community to collectively create high-quality, diverse, and inclusive datasets for machine learning.