Hugging Face

A developer spent two weeks optimizing a homelab with four RTX 3090s (96GB VRAM) for local LLM inference, achieving improvements like 40% throughput gain and 4x VRAM savings, but ultimately found that…

Alibaba's Qwen team released Qwen3-30B-A3B, a Mixture-of-Experts model with 30.5 billion total parameters but only 3.3 billion active per token, enabling it to run on a single 24 GB graphics card at s…

A developer released NeuroImprint Detector, an open-source framework that audits PEFT adapters in Federated Learning for backdoors that memorize training data. The tool can detect and reconstruct up t…

Hugging Face Spaces users report that their spaces are being auto-paused after 15 minutes of activity, with no logs and no ability to restart, forcing deletion and recreation. One user's space has bee…

A new simulation called 'Wall Street of AI Agents' allows users to model a multi-agent AI trading firm using small language models, hosted on Hugging Face Spaces. The project demonstrates how lightwei…

A Hugging Face Space, RomanaQureshi/Hayat-AI-Hajj-Medical-Emergency-Guidance, has been stuck at 'Starting' since June 20, 2026, despite the app running correctly on 0.0.0.0:7860 and the health checker…

A developer created two terminal commands that accept voice input for generating shell commands and answering questions using local speech-to-text and the Pi coding agent. The setup uses the hns CLI t…

A Docker Space on Hugging Face is stuck in APP_STARTING state with a stale runtime SHA, despite the app being reachable on port 7860. The issue persists after restarts and factory reboots, suggesting …

Egoist Machines, Inc. launched LodeDB, an open-source embedded vector database for local retrieval-augmented generation (RAG) that runs the same on-disk index on GPU when available, achieving up to 50…

A developer built a private mental health assistant using Llama 3 and Apple MLX that runs entirely on a MacBook, ensuring no data leaves the device. The system uses a 4-bit quantized version of Llama …

Modal Labs released state-of-the-art DFlash speculators for Qwen 3.5 and Qwen 3.6 models on Hugging Face, achieving 5-20% additional speedups and enabling Qwen 3.5 122B-A10B to run at over 1000 tok/s …

A developer released PaneTrans, a browser extension for drag-select region translation and OCR on video/canvas, built on Transformers.js with local processing by default. The tool uses an offscreen do…

Empero AI released Qwythos-9B-Claude-Mythos-5-1M, a 9-billion-parameter open-weights reasoning model distilled from Claude Mythos 5, featuring a 1-million-token context window, native tool use, and a …

Modal and Z Lab released DFlash, a speculative decoding model for Qwen 3.5 397B-A17B, achieving over 4.3x throughput versus baseline and 1.5x versus MTP on HumanEval at concurrency 1. The model uses a…

Microsoft, Hugging Face, Meta's PyTorch team, NVIDIA, and others launched OpenEnv, an open protocol for agent learning environments that standardizes how agents practice and improve. The protocol aims…

Headroom, an open-source local context compression layer, reduces LLM prompt sizes by 60-95% while preserving accuracy, solving cost and latency issues from massive context windows. Its multi-modal pi…

A developer building a fashion-discovery feed found that off-the-shelf AI image detectors perform poorly on custom data, achieving only 0.68 AUC versus a simple linear model's 0.82. The article argues…

Microsoft rolled out AI Citation Share and other features in Bing Webmaster Tools, while new data from Google and Ahrefs cast doubt on the effectiveness of llms.txt for AI search visibility. Google, M…

Liquid AI released two new retrieval models, LFM2.5-Embedding-350M and LFM2.5-ColBERT-350M, for fast multilingual search across 11 languages. The models, available on Hugging Face, are designed as dro…

Researchers at AIS2C2 2025 presented a prompt-driven tool-calling framework enabling lightweight open-source LLMs to perform complex multi-step tasks, reducing reliance on large proprietary models and…