Hugging Face

77 articles in the TruthFoundry index. Each links out to the original.

• IBM Releases Granite Switch 4.1.1 Patch With Updated AI Models · IBM launches backwards compatible Granite Switch 4.1.1 patch with improved ASR and adapter models.
• LiquidAI launches LFM3 edge-optimized multimodal foundation model family · LiquidAI releases LFM3, a 2B-parameter multimodal foundation model built for edge deployment.
• Ai2 study finds LLM research novelty judges are highly unreliable · Ai2 releases study showing LLM novelty judges produce inconsistent results with small setup changes.
• Researchers introduce LitPivot AI tool for research idea development · LitPivot system helps researchers refine ideas using contextual literature alignment.
• AWS Trainium model compatibility lookup tool released for Hugging Face · Public Hugging Face Space verifies Transformers model compatibility with AWS Trainium hardware.
• Microsoft and Hugging Face release ThinkingBox RL agent training on GitHub · ThinkingBox uses database state checks as RL rewards to improve agent reliability.
• OlmoEarth v1.3 open global satellite embedding model released · OlmoEarth launches v1.3 satellite embeddings with free global data and browser explorer
• IBM Releases Granite Time-Series Ensemble Forecasting System · IBM launches open Granite time-series ensemble to eliminate manual forecasting model selection.
• NVIDIA Kumo Relational outperforms supervised ML without task training · NVIDIA releases Kumo Relational pre-trained model for relational database predictions
• VocabX method cuts SLM inference cost 77% via domain vocabulary specialization · Researchers reduced production SLM inference cost by 77% optimizing vocabulary for actual inference cost.
• Tutorial: ML Profiling on AWS Trainium with TorchNeuron · Guide teaches system-level ML performance profiling for AWS Trainium hardware.
• IBM Releases Updated Granite Speech 5.0 AI Transcription Models · IBM launches three improved Granite Speech AI models for transcription, translation and speaker attribution.
• NVIDIA Unveils Cosmos3-H-Dreams for Real-Time Surgical Simulation · NVIDIA's Cosmos3-H-Dreams offers real-time generative simulation for clinical surgery, advancing interactive surgical training.
• FairStress benchmark exposes hidden fairness asymmetries in large language models · Researchers launch FairStress benchmark to measure hidden demographic fairness in LLM decision making.
• ALTK-Evolve tops τ²-bench leaderboard using learned memory with GPT-4.1 · ALTK-Evolve uses learned task guidelines to lead the τ²-bench customer service agent benchmark.
• AnyLanguageModel 1.0 Released For Apple Platforms After One Year · AnyLanguageModel 1.0 launches with unified API for local and hosted AI models on Apple platforms.
• Hugging Face launches native PyTorch TPU support via TorchTPU · Hugging Face ecosystem now natively runs on Google TPUs using new TorchTPU backend.
• PyTorch 2.15 Delivers Major MPS Performance Improvements For Mac · PyTorch 2.15 brings FP8 support and large speed gains for MPS on Apple Silicon Macs.
• TorchNeuron public beta launches 2026-10-20 for AWS Trainium Hugging Face models · TorchNeuron adds native PyTorch and Hugging Face support for AWS Trainium AI accelerators
• NVIDIA and AI4Bharat launch 2.8T token Indic pretraining corpus · NVIDIA and AI4Bharat announce 2.8 trillion token Indic language AI pretraining dataset.
• Reflection launches 501B open MoE LLM Beam on Hugging Face · US lab Reflection releases open 501B parameter MoE LLM Beam under Apache 2.0 license.
• Granite Switch 4.1 launches unified chat, decision, speech AI endpoint · Granite Switch 4.1 combines chat, decisions, and speech transcription on one hosted endpoint.
• Apertus 1.5 Open Multimodal AI Model Officially Released · Apertus 1.5 open AI models launched 2026-07-24 with multimodal and extended context features.
• Black Forest Labs releases FLUX.3 Image Klein AI model with Diffusers support · Black Forest Labs launches FLUX.3 Image Klein AI image model with native layout control.
• Meta releases Muse Spark Open 816B open-weight multimodal MoE AI model · Meta launches Muse Spark Open, an 816B parameter open-weight coding-optimized multimodal AI model.
• HiRO-ACEv2 Simulates Extreme Weather at 3 km Resolution for 1/300th the Cost · Researchers introduce HiRO-ACEv2, an AI system generating global 3 km climate data at 1/300th the cost of X-SHiELD.
• LiquidAI Launches LFM2.5-24B-A2B for Fast On-Device Agents · LiquidAI releases LFM2.5-24B-A2B, a 24B-parameter reasoning model optimized for efficient on-device agent execution.
• Ai2 replaces GPU cluster scheduler with budget-based fair share system · Allen Institute for AI deployed a budget-based GPU scheduler to improve research impact.
• Developer builds custom AI models using autonomous ML agent · A developer created seven custom AI models in days using an autonomous agent that plans, budgets, and trains on Hugging Face hardware.
• How to Unlock and Recover a Locked AOL Email Account · This guide explains why AOL accounts lock and provides steps to recover access.