✷ The archive · 11 labs · 871 dispatches
The labs
No more opening a dozen official blogs every morning. First-hand releases from OpenAI, Anthropic, DeepMind and the rest, each with its substance pulled out.
Vision Language Models Explained
Hugging Face provides a comprehensive guide to Vision Language Models (VLMs), detailing their architecture, open-source options, evaluation benchmarks, and new experimental support for fine-tuning via the TRL library.
Hugging Face and Google Cloud Vertex AI Model Garden Integration
Hugging Face has launched 'Deploy on Google Cloud,' enabling users to deploy thousands of open foundation models to Vertex AI or Google Kubernetes Engine (GKE) via the Hugging Face Hub or Vertex Model Garden.
CodeGemma Release Notes / What's New
Google has released CodeGemma, a family of open-access code-specialist LLMs based on Gemma, trained on 500 billion additional tokens of code, mathematics, and English language data.
Hugging Face Public Policy Program Overview and Submitted Materials
Hugging Face announced a comprehensive public‑policy program that provides U.S., EU, and U.K. policymakers with detailed position papers, testimony, and comment letters, reflecting its cross‑functional commitment to responsible openness and shaping AI regulation.
Hugging Face and Wiz Research Partnership for AI Security
Hugging Face has partnered with Wiz to integrate advanced vulnerability management and cloud security posture management to protect its platform and the broader AI/ML ecosystem.
Text2SQL with Hugging Face Dataset Viewer API and DuckDB-NSQL-7B
Hugging Face demonstrates how to use the DuckDB-NSQL-7B model and the Dataset Viewer API to convert natural language questions into SQL queries for analyzing over 120,000 open datasets.
SetFit Inference Acceleration with 🤗 Optimum Intel on Xeon
Hugging Face demonstrates how to achieve up to 7.8x faster inference throughput for SetFit models on Intel Xeon CPUs using post-training static quantization via the 🤗 Optimum Intel library.
Hugging Face and Cloudflare Workers AI Integration
Hugging Face integrated Cloudflare Workers AI to provide serverless GPU inference for popular open models, allowing developers to deploy AI applications with a pay-per-request pricing model.
Pollen-Vision: Unified Interface for Zero-Shot Vision Models in Robotics
Hugging Face and the Pollen Robotics team have released pollen-vision, an open-source library that integrates zero-shot vision models to enable robots to detect and localize unknown objects in 3D space.
Hugging Face Transformers: A Beginner's Guide to Open-Source ML
Hugging Face provides a comprehensive introductory guide to using the Transformers library and Hub to deploy and run open-source machine learning models like Microsoft's Phi-2.
Embedding Quantization: Binary and Scalar Techniques for Faster, Cheaper Retrieval
Hugging Face announced binary and int8 embedding quantization, cutting memory by 32× or 4× and speeding up retrieval up to 45× while keeping 96%–99% of original performance.
Hugging Face and Lighthouz AI Introduce Chatbot Guardrails Arena
Hugging Face and Lighthouz AI have launched the Chatbot Guardrails Arena, a community-driven stress-testing platform designed to evaluate the data privacy and security of LLMs and their guardrails.
GaLore: Advancing Large Model Training on Consumer-grade Hardware
GaLore enables the training of billion-parameter models on consumer-grade GPUs by reducing optimizer state memory requirements by over 82.5% through low-rank gradient projection.
Cosmopedia: Large-Scale Synthetic Data for LLM Pre-training
Hugging Face introduces Cosmopedia, the largest open synthetic dataset for LLM pre-training, containing 30 million files and 25 billion tokens generated by Mixtral-8x7B-Instruct-v0.1.
Phi-2 on Intel Meteor Lake: Local LLM Inference
Hugging Face demonstrates how to run the Microsoft Phi-2 model locally on Intel Meteor Lake (Core Ultra) processors using 4-bit quantization via OpenVINO and Optimum Intel.
Quanto: a PyTorch quantization backend for Optimum
Hugging Face introduces Quanto, a versatile and device-agnostic PyTorch quantization backend for Optimum designed to simplify low-precision model deployment across any modality.
Hugging Face Train on DGX Cloud
Hugging Face launched Train on DGX Cloud, a no-code service for Enterprise Hub organizations to fine-tune open models using NVIDIA H100 and L40S GPUs.
WebSight Dataset and Sightseer Model
Hugging Face introduced WebSight, a synthetic dataset of 2 million screenshot-to-HTML pairs, and Sightseer, a vision-language model capable of converting web screenshots into functional HTML code.
CPU Optimized Embeddings with Optimum Intel and fastRAG
Hugging Face and Intel have introduced a method to accelerate embedding models on Xeon CPUs using Optimum Intel and fastRAG, achieving up to 4.5x latency reduction and 4x throughput improvement via int8 quantization.
ConTextual: Benchmarking Multimodal Reasoning in Text-Rich Scenes
Hugging Face and UCLA researchers have introduced ConTextual, a dataset and leaderboard designed to evaluate how Large Multimodal Models (LMMs) jointly reason over text and visual cues in complex, text-rich images.
Hugging Face and Argilla Enable Collective Community Dataset Building
Hugging Face and Argilla have introduced a streamlined workflow using Hugging Face Spaces and Argilla to allow communities to collectively build high-quality, open-source datasets.
Text-Generation Pipeline on Intel Gaudi 2 AI Accelerator
Hugging Face introduces a custom text-generation pipeline for Intel Gaudi 2 AI accelerators, enabling streamlined deployment of Llama 2 models (7b, 13b, and 70b) via Optimum Habana.
StarCoder2 and The Stack v2 Release
BigCode has released StarCoder2, a family of transparently trained open code LLMs in 3B, 7B, and 15B parameter sizes, powered by the massive new Stack v2 dataset.
Hugging Face TTS Arena: Benchmarking Text-to-Speech Models
Hugging Face has launched the TTS Arena, a crowdsourced, side-by-side comparison tool and leaderboard using an Elo rating system to objectively measure text-to-speech model quality.
AI Watermarking 101: Tools and Techniques
Hugging Face provides a comprehensive overview of AI watermarking techniques across images, text, and audio to combat deepfakes and ensure content provenance.
Introduction to Matryoshka Embedding Models
Matryoshka Embedding models allow for variable-size embeddings that can be truncated without significant performance loss, enabling a flexible trade-off between storage, speed, and accuracy.
Fine-Tuning Gemma Models in Hugging Face
Hugging Face provides a guide on using Parameter-Efficient Fine-Tuning (PEFT) and Low-Rank Adaptation (LoRA) to customize Google's Gemma models on GPUs and Cloud TPUs.
Hugging Face and Haize Labs Introduce Red-Teaming Resistance Leaderboard
Hugging Face and Haize Labs have launched the Red-Teaming Resistance (RTR) Benchmark to evaluate LLM robustness against high-quality, human-like adversarial prompts across specific safety violation categories.
Google Gemma Open LLM Release
Google has released Gemma, a family of open-access large language models based on Gemini, available in 2B and 7B parameter sizes with base and instruction-tuned variants.
Open Ko-LLM Leaderboard
Hugging Face and Upstage have launched the Open Ko-LLM Leaderboard to provide a fair, transparent evaluation ecosystem for Korean Large Language Models using private test sets to prevent contamination.
Hugging Face PEFT New LoRA Merging Methods
Hugging Face has introduced several new merging methods to the PEFT library, enabling users to combine multiple LoRA adapters from the same base model on the fly to synthesize new capabilities.
Synthetic Data with Open-Source LLMs Cuts Cost, Latency, and Carbon for Custom Models
Hugging Face shows how using open‑source LLMs to generate synthetic data and fine‑tune a small RoBERTa model reduces inference cost from $3061 to $2.7, latency from seconds to 0.13 s, and CO₂ emissions from ~1 t to 0.12 kg while matching GPT‑4 accuracy on financial sentiment classification.
AMD Pervasive AI Developer Contest
AMD and Hugging Face have partnered to launch the Pervasive AI Developer Contest, offering developers free access to AMD hardware and cash prizes to build AI applications in Generative AI, Robotics AI, and PC AI.
Hugging Face TGI Messages API Release
Hugging Face has introduced a Messages API for Text Generation Inference (TGI) starting with version 1.4.0, enabling OpenAI Chat Completion API compatibility for open LLMs.
SegMoE: Segmind Mixture of Diffusion Experts
SegMoE is a framework for creating Mixture-of-Experts (MoE) Diffusion models by replacing Feed-Forward layers in Stable Diffusion architectures with sparse MoE layers to improve prompt understanding.
NPHardEval Leaderboard: Evaluating LLM Reasoning via Computational Complexity
Hugging Face introduces the NPHardEval leaderboard, a dynamic benchmark that uses computational complexity classes to quantitatively measure the logical reasoning abilities of Large Language Models.
PatchTST Integration in Hugging Face
Hugging Face has integrated PatchTST, a Transformer-based model that uses time series patching and channel-independence to improve long-term forecasting and enable transfer learning.
Hugging Face Text Generation Inference now supports AWS Inferentia2
Hugging Face announced the general availability of Text Generation Inference on AWS Inferentia2 via Amazon SageMaker, enabling cost‑effective, high‑throughput LLM serving as an alternative to GPU deployments.
Constitutional AI with Open LLMs
Hugging Face introduces an end-to-end recipe and the llm-swarm tool to implement Constitutional AI (CAI) on open models, enabling scalable self-alignment based on user-defined principles without expensive human feedback.
Enterprise Scenarios Leaderboard: Evaluating LLMs for Real-World Use Cases
Hugging Face and Patronus AI have launched the Enterprise Scenarios Leaderboard to evaluate language models on six real-world business tasks, moving beyond academic benchmarks to measure practical enterprise utility.
Accelerating StarCoder on Intel Xeon with Optimum Intel
Hugging Face and Intel demonstrate over 7x inference acceleration for the StarCoder-15B model on 4th Gen Intel Xeon processors by combining 8-bit quantization and assisted generation.
Hugging Face Hallucinations Leaderboard launch and initial findings
Hugging Face launched the Hallucinations Leaderboard to benchmark LLMs on factuality and faithfulness errors across multiple open-source datasets, offering transparent rankings that guide model selection and research.
AI Secure LLM Safety Leaderboard
Hugging Face and the Secure Learning Lab have released the LLM Safety Leaderboard, powered by the DecodingTrust framework to evaluate LLM trustworthiness across eight critical safety dimensions.
Hugging Face and Google Cloud Strategic Partnership
Hugging Face and Google Cloud have entered a strategic partnership to democratize machine learning by integrating open models with Google Cloud's AI infrastructure and hardware.
Open-source LLMs as LangChain Agents
Hugging Face demonstrates that open-source LLMs, specifically Mixtral-8x7B, are now capable of powering agent workflows and can outperform GPT-3.5 in general-purpose reasoning tasks.
Fine-Tuning Wav2Vec2-BERT for Low-Resource ASR
Hugging Face demonstrates how to fine-tune Meta's Wav2Vec2-BERT model for Automatic Speech Recognition (ASR) in low-resource languages, achieving performance comparable to Whisper-large-v3 while being significantly faster and more resource-efficient.
PatchTSMixer added to Hugging Face Transformers – release and quick‑start guide
PatchTSMixer, a lightweight MLP‑Mixer time‑series model from IBM Research, is now released in Hugging Face Transformers, offering state‑of‑the‑art forecasting with far lower memory and runtime costs.
Preference Tuning LLMs with Direct Preference Optimization Methods – Empirical Comparison of DPO, IPO, and KTO
Hugging Face evaluated DPO, IPO and KTO alignment methods on two 7B chat models, showing DPO consistently outperforms the others when the beta hyper‑parameter is properly tuned.
Accelerating SD Turbo and SDXL Turbo Inference with ONNX Runtime and Olive
Hugging Face and Microsoft introduce optimizations using ONNX Runtime and Olive to achieve throughput gains up to 229% for SDXL Turbo and 120% for SD Turbo compared to PyTorch.
Run ComfyUI Workflows on Hugging Face Spaces with Gradio
Hugging Face provides a guide to converting complex ComfyUI workflows into Gradio applications for free, serverless deployment on Hugging Face Spaces ZeroGPU.