Selected industry news from AIHOT. Follow advances in AI, products and technology.
AIHOT · Selected summaries with original sources · Nov 4, 2026, 9:00 AM
Hugging Face officially announced the launch of the Train on DGX Cloud service, available to Enterprise Hub organizations, supporting the fine-tuning of Llama using H100 or L40S GPUs on NVIDIA DGX Cloud.
Hugging Face has released Quanto, a PyTorch quantization backend that supports int2/int4/int8/float8 weights and int8/float8 activations. It runs on CPU/GPU/MPS devices, is compatible with torch.compile, and has been integrated into the transformers library, accessible directly via QuantoConfig.
Hugging Face released the WebSight dataset. Version 0.1 contains 823,000 pairs of synthetic screenshot/HTML data. Version 0.2 was upgraded to 2 million pairs, using real screenshots and switching to Tailwind CSS. The Sightseer model, fine-tuned on this dataset, can convert webpage screenshots into HTML code with matching images. Resources include the dataset, a technical report (arXiv:2403.09029), and Colab examples.
Hugging Face and Intel jointly released CPU-optimized embedding models based on Optimum Intel and IPEX (such as int8 static quantized versions of BGE-small/base/large).
Hugging Face, in collaboration with researchers from the University of California, Los Angeles (UCLA), released the ConTextual dataset, which contains 506 instructions requiring joint reasoning based on image and text cues in a context-sensitive manner, covering time reading.
Hugging Face and Argilla have launched a collaborative dataset building workflow based on the integration of Spaces and Argilla, allowing users to participate in annotation within seconds using their Hugging Face accounts. This workflow has been used to generate the 10k_prompts_ranked dataset, which contains 10,000 prompts with user ratings. The first community co-building program is now open, offering free storage, enhanced CPU Spaces, and joint promotion support.
Hugging Face published a blog post announcing support for Intel® Gaudi® 2 AI accelerators in its text generation pipeline, compatible with Llama-2-7b, 13b, and 70b models.
BigCode has released the StarCoder2 series of open-source code large language models, available in three sizes: 3B, 7B, and 15B, all trained on the newly released The Stack v2 dataset.
Hugging Face has launched TTS Arena, a text-to-speech model evaluation platform inspired by LMSys Chatbot Arena. It allows users to submit text, listen to speech generated by two models, and vote for the more natural one; results are compiled into a public leaderboard using an Elo-like algorithm. Initial models include ElevenLabs, MetaVoice, OpenVoice, Pheme, WhisperSpeech, and XTTS.
Mistral AI has released its conversational assistant, le Chat, which supports Mistral Large, Mistral Small, and the prototype model Mistral Next. It offers multilingual interaction and system-level adjustable moderation mechanisms, while simultaneously launching a self-deployed version for enterprises, le Chat Enterprise.
Mistral AI has released its flagship large language model, Mistral Large, which supports a 32K context window, native multilingual capabilities (English, French, Spanish, German, Italian), function calling, and JSON format output. It is now available via la Plateforme and Azure AI Studio.
The Hugging Face blog systematically introduces watermarking techniques for AI-generated content, covering the principles of visible and invisible watermarks, and comparing embedding during generation (e.g., IMATAG, Stable Diffusion XL Turbo integration) with post-generation addition (e.g., Truepic).
Hugging Face released a tutorial introducing how to perform parameter-efficient fine-tuning (PEFT) on Google DeepMind's open-source Gemma (2B/7B) models on GPUs and Cloud TPUs.
Haize Labs, in collaboration with Hugging Face, released the Red-Teaming Resistance Benchmark (RTR), which systematically evaluates the robustness of frontier large language models against high-quality adversarial prompts constructed by humans.
The Hugging Face blog systematically introduces Matryoshka embedding models, explaining their characteristic of concentrating key information in the leading dimensions of embedding vectors to support arbitrary truncation. It also describes how to train (using MatryoshkaLoss) and perform inference (using the truncate_dim parameter) with the Sentence Transformers framework.
Google releases the open-source large language model Gemma, including four open model variants: gemma-2b, gemma-2b-it, gemma-7b, and gemma-7b-it, with a context window of 8K tokens.
Upstage and Hugging Face jointly launched the Open Ko-LLM Leaderboard, an open evaluation platform designed specifically for Korean large language models, which includes Ko-ARC.
Hugging Face has added support for multiple dynamic LoRA adapter merging methods in the PEFT library, including concat, linear, svd, ties, dare, and magnitude_prune.
A Hugging Face blog post introduces how to use open-source LLMs such as Mixtral-8x7B-Instruct-v0.1 to generate synthetic data, training models that achieve comparable performance to GPT4 (94% accuracy) on investor sentiment recognition tasks.
OpenAI explores training generative models on large-scale video data, jointly training text-conditioned diffusion models on videos and images with variable durations, resolutions, and aspect ratios; employing a Transformer architecture operating on spatiotemporal patches of video and image latent codes; the largest model, Sora, can generate one-minute high-fidelity videos; results indicate that scaling video generation models is a viable path toward building general-purpose simulators of the physical world.
OpenAI is testing ChatGPT's memory feature, enabling it to remember content discussed by users to improve the helpfulness of subsequent conversations. Users have full control over this memory.
Hugging Face introduced the Messages API in Text Generation Inference (TGI) version 1.4.0, offering an interface compatible with the OpenAI Chat Completion API.
SegMoE is a new framework for building Mixture-of-Experts (MoE) diffusion models from scratch. It has been deeply integrated into the Hugging Face ecosystem and supports diffusers; this release includes SegMoE-4x2.
The Hugging Face blog announced the launch of the NPHardEval leaderboard. This benchmark was developed by researchers from the University of Michigan and Rutgers University. It includes 900 algorithmic problems covering P, NP-complete, and NP-hard complexity classes, with automatic monthly updates. The evaluation uses dual metrics: Weighted Accuracy (WA) and Failure Rate (FR). All problems are automatically generated and verified based on computable problems, excluding numerical calculations to focus on logical reasoning. The code, dataset, and leaderboard have been open-sourced on GitHub and Hugging Face.