Engineering
PRX Part 4: Our Data Strategy
Photoroom describes the data pipeline behind its PRX text-to-image model, assembling training data from public and internal sources while prioritizing…
Engineering
Featuring Every Eval Ever Results on Hugging Face Model Pages
The Every Eval Ever (EEE) project and Hugging Face Community Evals have achieved interoperability, letting researchers submit evaluation results to…
Engineering
DiScoFormer: One Transformer for Density and Score, Across Distributions
DiScoFormer is a transformer-based architecture that estimates both the density and score of a data distribution in a single forward…
Engineering
ScarfBench: Benchmarking AI Agents for Enterprise Java Framework Migration
IBM Research introduced ScarfBench, an open benchmark evaluating AI agents on enterprise Java migration tasks across Spring, Jakarta EE, and…
Engineering
Tuning the Harness, Not the Model: A Nemotron 3 Ultra Playbook
LangChain optimized NVIDIA’s open Nemotron 3 Ultra model for agent tasks by tuning the harness — system prompts, tool descriptions,…
Engineering
Deep Agents Code on NemoClaw: A Governed Blueprint for Your Most Sensitive Code
LangChain released Deep Agents Code integrated with NVIDIA’s NemoClaw blueprint to run coding agents securely against sensitive codebases. The system…
Engineering
How Schneider Electric Built Their LLMOps Foundations at Enterprise Scale With LangSmith
Schneider Electric built its LLMOps infrastructure on a self-hosted LangSmith deployment to manage more than 60 AI agents across the…
Engineering
vLLM x HPC-Ops: High-Performance Attention and MoE Backends from Tencent Hunyuan
Tencent Hunyuan’s HPC-Ops operator library contributed two high-performance kernels to vLLM targeting Hopper GPUs. The attention backend uses dynamic load-balanced…
Engineering
Experience and Lessons Learned from Serving Multi-Stage Qwen3-Omni in vLLM-Omni
The vLLM-Omni team details how it serves Qwen3-Omni, a multimodal text-and-speech model, using a three-stage pipeline of Thinker (reasoning), Talker…
Engineering
Engineering TTS Inference in vLLM-Omni
The vLLM-Omni team detailed model-specific engineering work to optimize text-to-speech inference across four models. For Qwen3-TTS the team decoupled streaming…
Engineering
Beyond One Model: Fusion in vLLM Semantic Router
vLLM Semantic Router introduced Fusion, a routing primitive that runs multiple models concurrently, uses a judge model to analyze agreement…
Engineering
Profiling in PyTorch (Part 3): Attention is all you profile
The third installment in Hugging Face’s “Profiling in PyTorch” series examines how different attention mechanisms perform under GPU profilers. The…