Hugging Face engineers reproduced a viral AI art project in which a language model trained with reinforcement learning writes JavaScript to paint watercolours, publishing all artifacts openly. The pro…
Hugging Face released funes, a durable memory layer for coding agents that indexes and retrieves session traces locally, supporting Claude Code, Codex, pi, and Hermes. The tool, available as a single …
Fine-tuning LiquidAI's LFM2.5-350M model with Group Relative Policy Optimization (GRPO) using the TRL library improved its IFStruct benchmark score from 22.6% to 29.7% in just 100 training steps, acco…
A developer has released AMB, an open and preregistered benchmark that tests whether memory layers in coding agents improve real coding work, using executable tests to grade artifacts in a sandboxed r…
A new inquiry asks whether the datasets required for AI models, products, or research objectives are actually available, highlighting gaps in ground truth, population coverage, language, geography, an…
A Hugging Face forum user asked which models are suitable for building an AI chatbot, with related discussions covering model selection for projects, open-source LLM choices for chatbot development, t…
Multiple full Vision Transformer (ViT) object detection models exist, including LW-DETR, RF-DETR, YOLOS, ViTDet, and WB-DETR, each differing in architectural purity and usability. The choice depends o…
A user identified as Adam0111 reported on the Hugging Face Forums on September 2, 2026, that they cannot use the BAAI/bge-m3 model via the Hugging Face Inference API. The post, titled 'BAAI/bge-m3 iss…
Hugging Face's public progress checker for the Audio Course is failing server-side before evaluating usernames, due to an inability to access the Unit 7 pass roster from 'huggingface-course/audio-cour…
A researcher's causal experiments on nanoGPT, ResNet-18, and a diffusion U-Net indicate that neural networks behave as classical stateful dynamical systems, where the same update has different effects…
The Allen Institute for AI introduced BenchMIRT, a new method for auditing LLM benchmarks at the level of individual prompts, which uses multidimensional Item Response Theory to separate the underlyin…
Model weights are the adjustable numerical parameters inside a machine learning model that are fine-tuned during training to improve performance, analogous to ingredient amounts in a recipe. Downloadi…
Hugging Face's Datasets 5.0.0 changed IterableDataset.shuffle() to fill the shuffle buffer from multiple input shards, which collapses an 8-shard Parquet stream into 1 logical shard and leaves only 1 …
A causal investigation of vocabulary-logit steering on Qwen2-1.5B found that the largest coordinates are not necessarily the ones driving the effect, with transport appearing around K≈150 and persisti…
Hugging Face user Nargan requested the unpausing of their legacy Space 'Nargan/projects', which was automatically suspended and flagged as abusive (Error 503) due to a code bug causing excessive polli…
Hugging Face's paper submission pipeline currently only accepts arXiv identifiers, rejecting Zenodo records such as zenodo.19848279, according to a feature request from independent researcher Massimo.…
Hugging Face released @huggingface/kernels, a JavaScript library for loading and running optimized WebGPU kernels from the Hugging Face Hub, along with an initial collection of 207 kernels published a…
Hugging Face forum user dpizza reported that the WhisperForAudioClassification class from the transformers library ignores the num_labels argument when loading openai/whisper-tiny, leaving the classif…
In a Q&A on fine-tuning large language models, Sahar advises using the Instruct version of models like Llama 3.1 8B, Gemma 2 9B, and Mistral 7B for most tasks, including multiple-text classification, …
Varshith Reddy, an independent student and open-source developer, has requested a Hugging Face Community Hardware Grant to host ARIA (AI-Powered Repository Intelligence Agent), an MIT-licensed develop…