Sakhanda Wire
NVDA $153.55 -1.66% MSFT $504.27 -2.65% GOOGL $281.47 -0.79% META $620.64 -0.92% AMZN $256.56 -0.46%

Hugging Face

111 stories · all sources

Hugging Face

Isaac Sim walkthrough demonstrates SimReady robot pick-and-place

A new walkthrough presents a beverage-can pick-and-place task in Isaac Sim using the UR17G arm, Robotiq 2F-140 gripper and a beverage can. The assets are assembled into one scene, validated, and controlled to grasp the can, lift it, move it to a target and release it. The example uses Isaac Sim 6.2 and an agent to generate task controls; GPT-6 Sol is used in the walkthrough, although other agent and model combinations may also work.

Hugging Face

XYBench tests whether AI assistants solve the problem behind the question

Researchers are releasing XYBench, a benchmark containing 8,115 real-world questions designed to test whether AI assistants can recognize mistaken assumptions and address users’ underlying goals. The dataset covers programming and everyday activities, including cooking and home repair, and received the Best Resource Paper Award at EMNLP 2026.

Hugging Face

llama.cpp and ggml reach version 1.0.0

The llama.cpp project has released version 1.0.0 of its libraries, alongside version 1.0.0 of ggml, the machine-learning tensor library. The team said the milestone reflects the growing number of downstream applications that rely on their interfaces. From now on, incompatible changes to the public API or ABI in llama.cpp and ggml will be marked by major-version increases.

Hugging Face

IBM researchers release LLMON previews for Granite 4.1 models

IBM researchers have released two experimental preview models, granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview. They are based on Granite 4.1 base checkpoints and are trained to read and generate LLMON, or LLM Object Notation, a markup format designed for large language models.

Hugging Face

Guide details LoRA fine-tuning for TRELLIS.2 3D generation

A new guide describes scripts and Docker containers for adapting TRELLIS.2, a 3D generation model, to specialized asset collections. The workflow can run on local hardware or through Hugging Face Jobs, provided the dataset’s meshes are uploaded to the Hugging Face Hub. The guide focuses on improving shoe geometry using 188 footwear objects selected from the 1,030-object Google Scanned Objects dataset.

Hugging Face

BotPsy brings a 0.8B-parameter agent to on-device assistants

Tether AI Research has introduced BotPsy, a 0.8B-parameter language model built on Qwen3.5-0.8B for function calling and agentic workflows. The model is intended for on-device personal-assistant tasks including web research, travel, commerce, finance, device control, file management, calendars, and messaging. Its developers say it is best suited to low- and medium-difficulty tasks and is not designed for high-stakes use.

Hugging Face

Pooling-aware training improves compressed multi-vector retrieval

Multi-vector embedding models represent a document with separate contextual vectors for its tokens. This can improve search because different query terms may match different document passages, but the approach requires substantial storage. Token pooling reduces that cost by merging similar vectors. HierarchicalTokenPooling uses Ward clustering and represents each resulting group by its mean vector.

Hugging Face

Hugging Face adds Google Cloud billing for Hub usage

Hugging Face organizations can now connect their accounts to Google Cloud through the Google Cloud Marketplace. Once linked, charges for Hugging Face usage appear on the organization’s existing Google Cloud invoice rather than being charged to a credit card.

Hugging Face

Hugging Face presents decision models for structured AI choices

Hugging Face has published an overview of decision models, an approach intended for applications that need structured choices rather than generated text. Given a state, a set of questions and predefined answers, such a model returns decisions together with probability distributions. The article presents the interface as “State + Questions + Allowed Answers → Decisions + Probabilities.”

Hugging Face

Synthetic dataset maps Japanese demographics and household finances

Elith and NVIDIA have released Nemotron-Personas-Finance-Japan, a synthetic dataset designed for financial AI development in Japan. It combines official demographic and labor statistics with information on household finances and financial literacy, allowing teams to create and assess systems without using real customer records. The dataset is available for commercial and non-commercial use under the CC BY 4.0 license.

Hugging Face

HOPE prunes MoE experts while preserving cooperative groups

A method called HOPE aims to reduce the memory footprint of mixture-of-experts (MoE) language models by removing less useful experts while accounting for relationships between them. MoE models may contain hundreds of experts, but only a small subset is activated for each token. In Qwen3.5-122B, for example, 256 experts are present in each of 48 MoE layers, while eight are used per layer for an individual token. Expert MLPs account for about 95% of the model’s roughly 250 GB parameter memory footprint.