Tag
Modal 宣布达成 10 亿美元年化营收并完成 D 轮融资，同时继续推广其面向 AI 工作负载的容器化云基础设施平台，涵盖推理、训练、沙盒与函数计算。
DatologyAI 发布 Curation Studio，面向训练自研模型的团队提供可定制的前沿数据筛选服务，并在 Modal 上利用数百块 B300 GPU 进行大规模合成数据测试验证。
Aleph Alpha describes how they scaled pre-training of their 30B-A3B MoE model (revealed to be Kolibri Origin) from 16 to 512 NVIDIA B200 GPUs using a hierarchical hyperparameter search approach, achieving 35.3% MFU and near-linear scaling just 6% below ideal, trained with a custom fork of torchtitan over InfiniBand/NVLink.
Benchmarking distributed LLM inference on Kubernetes reveals unauthenticated control-plane ports and security exposures, with mitigations like NetworkPolicy and WireGuard impacting performance.
The article promotes a presentation on Elastic Expert Parallelism in vLLM at the PyTorch Conference North America 2026, discussing how to dynamically add or remove GPUs in Mixture-of-Experts deployments with minimal downtime.
Modal details how they optimized inference performance for trillion-parameter coding agents, achieving significant improvements in throughput and interactivity for their service.
SGLang-Diffusion with VDN-H3 from MiniMax enables fast and scalable video generation, achieving 14.4s of 768p video in just 9.0s on 8× B200 GPUs for faster-than-real-time inference.
Modal is a serverless cloud platform designed for AI workloads, supporting inference, training, and sandboxes. It enables instant scaling from zero to thousands of GPUs with pure Python code, significantly reducing latency and accelerating time to market.
A user reports near-linear performance scaling when adding a second RTX 3090 for inference with a Qwen model, achieving roughly 1.8x decode TPS improvement without NVLink.
A Modal tutorial demonstrating how to scale protein binder design using ESMFold2 and ESMC models, with code for iterative optimization and autoscaling infrastructure.
OpenAI shares their deep learning infrastructure approach and open-sources kubernetes-ec2-autoscaler, a batch-optimized scaling manager for Kubernetes, emphasizing how infrastructure quality multiplies research progress.