We fit a full offline voice agent into 1.2 GB on Android
aufklarer
• • 1NEW Articles from Team or Enterprise organizations will get promoted to the main section.
Why I Started Building Machine Learning in Public
KissTheHabit
• Lucy 2.5 Real-Time AI Video Editing: Decart's 40ms Live Stream Revolution
ResterChed
• 🚀 Vedika: The Next Generation of Long-Context Open Models
Veda-Labs
• • 2Giving agent frameworks a threat-intel safety check: RelayShield's smolagents tools
relayshieldadmin
• • 1The Effects of AI on Human Cognition: The Principle of Computational Irreducibility
Bhavibond
• MLA in the Transformers modeling backend for vLLM
hmellor
• • 1The Transformers modeling backend is now as fast as native vLLM
hmellor
• • 1Fine-tune video and image models at scale with NVIDIA NeMo Automodel and 🤗 Diffusers
nvidia
• • 78When Activation Checkpointing Calls a Stateful Quantizer Twice
KissTheHabit
• How IDA’s First 156K-Parameter Judgment Family Governed DeepSeek in Production
KissTheHabit
• Kimi K3 Model Overview: 2.8T Parameters, MXFP4 Quantization, and What the Open Weights Mean for the Community
ResterChed
• • 75Beyond the Model Checkpoint: Building AI as a Governed Institution
KissTheHabit
• StripeCycle: A CycleGAN You Can Actually Read — From-Scratch PyTorch, Start to Finish
turhancan97
• • 1When will language models be good enough?
craffel
• • 5The Topology of Human Systems
Bhavibond
• Architecture trends over 5 years
embedl
• • 1SQRL Digs Before It Queries
feyninc
• NVIDIA Nemotron 3 Embed Ranks #1 Overall on RTEB, Advancing Agentic Retrieval
nvidia
• • 55One Adapter, Both Modalities: Field Notes from Building and Serving a Multimodal Reranker
lightonai
• • 16