---
title: "AlphaSignal Training Infra News"
url: "https://alphasignal.ai/topic/training-infra"
description: "Latest AI and machine learning news about Training Infra on AlphaSignal."
---

# AlphaSignal Training Infra News

Latest AI and machine learning news about Training Infra on AlphaSignal.

## Articles

- [Ai2 Rebuilt Its GPU Scheduler and Cut Wait Times by 87%](https://alphasignal.ai/news/ai2-rebuilt-its-gpu-scheduler-and-cut-wait-times-by-87.md) (2026-10-09)
- [UW and NVIDIA's SGS Trains Real Robots Across a Million Simulated Environments](https://alphasignal.ai/news/uw-and-nvidia-s-sgs-trains-real-robots-across-a-million-simulated-environments.md) (2026-10-08)
- [Mila's RoboJEPA Lets Robotics Teams Predict Performance Before Buying Compute](https://alphasignal.ai/news/mila-s-robojepa-lets-robotics-teams-predict-performance-before-buying-compute.md) (2026-10-07)
- [Zyphra Cuts MoE Training Communication Costs by 2.63x Using Router Patterns](https://alphasignal.ai/news/zyphra-cuts-moe-training-communication-costs-by-2-63x-using-router-patterns.md) (2026-10-07)
- [Optimizer Rankings Flip as Batch Size Grows, Study Warns ML Engineers](https://alphasignal.ai/news/optimizer-rankings-flip-as-batch-size-grows-study-warns-ml-engineers.md) (2026-10-06)
- [Stanford's EigenDEXplore Trains Robot Hands 82% Better Using Human Motion](https://alphasignal.ai/news/stanford-s-eigendexplore-trains-robot-hands-82-better-using-human-motion.md) (2026-10-06)
- [LoGRA Cuts AI Reinforcement Training Memory by 45% on One Node](https://alphasignal.ai/news/logra-cuts-ai-reinforcement-training-memory-by-45-on-one-node.md) (2026-10-05)
- [Yousef Saad's Survey Bridges Math and LLMs With One Shared Language](https://alphasignal.ai/news/yousef-saad-s-survey-bridges-math-and-llms-with-one-shared-language.md) (2026-10-03)
- [SmolLM3 Retrieves Perfectly at 64x Its Training Length by Moving Dropout](https://alphasignal.ai/news/smollm3-retrieves-perfectly-at-64x-its-training-length-by-moving-dropout.md) (2026-10-03)
- [Google Sends Trillium TPUs to Orbit to Power AI Beyond Earth's Grid](https://alphasignal.ai/news/google-sends-trillium-tpus-to-orbit-to-power-ai-beyond-earth-s-grid.md) (2026-10-02)
- [Google Moves Gboard's Private AI Training Into Secure Server Enclaves](https://alphasignal.ai/news/google-moves-gboard-s-private-ai-training-into-secure-server-enclaves.md) (2026-10-02)
- [Meta AI's NAVA-WAM Teaches Robots From Unlabeled Video, Hitting 93% Success](https://alphasignal.ai/news/meta-ai-s-nava-wam-teaches-robots-from-unlabeled-video-hitting-93-success.md) (2026-10-02)
- [Ai2 Rebuilds OLMo-Core 3 to Train Trillion-Parameter AI 2.7x Faster](https://alphasignal.ai/news/ai2-rebuilds-olmo-core-3-to-train-trillion-parameter-ai-2-7x-faster.md) (2026-10-01)
- [Boston Dynamics Gives Atlas a 13-Motion Hand Built for AI Training](https://alphasignal.ai/news/boston-dynamics-gives-atlas-a-13-motion-hand-built-for-ai-training.md) (2026-10-01)
- [LOOM Pushes MoE Transformers to 12 Loops, Boosting Accuracy by 5.3 Points](https://alphasignal.ai/news/loom-pushes-moe-transformers-to-12-loops-boosting-accuracy-by-5-3-points.md) (2026-10-01)
- [Looped Prediction Transformer Cuts AI Memory by 79% With One Shared Cache](https://alphasignal.ai/news/looped-prediction-transformer-cuts-ai-memory-by-79-with-one-shared-cache.md) (2026-10-01)
- [Pangram Labs Warns AI Web Text Could Triple Pretraining Costs by 2028](https://alphasignal.ai/news/pangram-labs-warns-ai-web-text-could-triple-pretraining-costs-by-2028.md) (2026-09-30)
- [Synthetic Warm-Up Saves 21B Training Tokens by Boosting Long-Range Retrieval](https://alphasignal.ai/news/synthetic-warm-up-saves-21b-training-tokens-by-boosting-long-range-retrieval.md) (2026-09-30)
- [ID Balancing Cuts Routing Overload by 89% in 70B Sparse AI Models](https://alphasignal.ai/news/id-balancing-cuts-routing-overload-by-89-in-70b-sparse-ai-models.md) (2026-09-30)
- [Stanford's SOMA Slashes AI Training Noise by Sharding Models Into Expert Shards](https://alphasignal.ai/news/stanford-s-soma-slashes-ai-training-noise-by-sharding-models-into-expert-shards.md) (2026-09-29)
- [DataTalks.Club's Free ML Zoomcamp Hits 14,600 Stars and Opens 2026 Registration](https://alphasignal.ai/news/datatalks-club-s-free-ml-zoomcamp-hits-14-600-stars-and-opens-2026-registration.md) (2026-09-29)
- [A Silent BF16 Bug in FlashAttention-3 Can Derail Entire Pretraining Runs](https://alphasignal.ai/news/a-silent-bf16-bug-in-flashattention-3-can-derail-entire-pretraining-runs.md) (2026-09-28)
- [Researchers Train AI on Zero Human Data Using Self-Play Programs](https://alphasignal.ai/news/researchers-train-ai-on-zero-human-data-using-self-play-programs.md) (2026-09-24)
- [Tencent Cuts AI Training Time by 29% With a Simple Batch Size Trick](https://alphasignal.ai/news/tencent-cuts-ai-training-time-by-29-with-a-simple-batch-size-trick.md) (2026-09-24)
- [Tsinghua's TactileStep Teaches Humanoid Robots to Land 48% Softer](https://alphasignal.ai/news/tsinghua-s-tactilestep-teaches-humanoid-robots-to-land-48-softer.md) (2026-09-24)
- [Prime Intellect Opens Prime Sandboxes to Run 30M AI Agent Environments](https://alphasignal.ai/news/prime-intellect-opens-prime-sandboxes-to-run-30m-ai-agent-environments.md) (2026-09-23)
- [White Circle's Halo Open-Source Framework Trains 20B Models 2.8x Faster](https://alphasignal.ai/news/white-circle-s-halo-open-source-framework-trains-20b-models-2-8x-faster.md) (2026-09-22)
- [Meta's Petal Cable Will Move 1 Petabit per Second Across the Atlantic](https://alphasignal.ai/news/meta-s-petal-cable-will-move-1-petabit-per-second-across-the-atlantic.md) (2026-09-21)
- [Østfold Researchers Show Tiny 9,800-Parameter AI Shares Hidden Features Across Classes](https://alphasignal.ai/news/stfold-researchers-show-tiny-9-800-parameter-ai-shares-hidden-features-across.md) (2026-09-18)
- [A 7.4B Model Matches GPT-3 13B Using 20x Less Compute](https://alphasignal.ai/news/a-7-4b-model-matches-gpt-3-13b-using-20x-less-compute.md) (2026-09-16)
- [Bojie Li's Open Textbook Teaches AI Infrastructure Through Hardware Limits](https://alphasignal.ai/news/bojie-li-s-open-textbook-teaches-ai-infrastructure-through-hardware-limits.md) (2026-09-16)
- [Sakana AI's PC-ALM Trains 1,000-Layer Networks Without Backpropagation](https://alphasignal.ai/news/sakana-ai-s-pc-alm-trains-1-000-layer-networks-without-backpropagation.md) (2026-09-14)
- [Open-Source Reef Turns Agent Feedback Into Live Model Updates Automatically](https://alphasignal.ai/news/open-source-reef-turns-agent-feedback-into-live-model-updates-automatically.md) (2026-09-12)
- [PyTorch 2.14 Ships Faster Apple Silicon and 100x GPU Speedups](https://alphasignal.ai/news/pytorch-2-14-ships-faster-apple-silicon-and-100x-gpu-speedups.md) (2026-09-11)
- [George Hotz's tinygrad Beats AMD's Best Llama 3.1 Training Time by 24 Seconds](https://alphasignal.ai/news/george-hotz-s-tinygrad-beats-amd-s-best-llama-3-1-training-time-by-24-seconds.md) (2026-09-09)
- [Magic Matches DeepSeek V4 Pro Base Using 50x Less Compute](https://alphasignal.ai/news/magic-matches-deepseek-v4-pro-base-using-50x-less-compute.md) (2026-09-08)
- [When to Gate Recursive Terminal-task Synthesis](https://alphasignal.ai/news/when-to-gate-recursive-terminal-task-synthesis.md) (2026-09-08)
- [Kyoto University Finds a Hard Speed Limit on What SGD Can Learn](https://alphasignal.ai/news/kyoto-university-finds-a-hard-speed-limit-on-what-sgd-can-learn.md) (2026-09-08)
- [Prime Intellect Slashes GLM-5.2 Weight Transfer by 22x With RDMA](https://alphasignal.ai/news/prime-intellect-slashes-glm-5-2-weight-transfer-by-22x-with-rdma.md) (2026-09-03)
- [PyTorch 2.14 Ships Faster Apple Silicon, Rebuilt Distributed Training and NVGEMM](https://alphasignal.ai/news/pytorch-2-14-ships-faster-apple-silicon-rebuilt-distributed-training-and-nvgemm.md) (2026-09-02)
- [UC Berkeley Team Cuts Diffusion Model Training Time by 40% With Frequency-Aware Loss](https://alphasignal.ai/news/uc-berkeley-team-cuts-diffusion-model-training-time-by-40-with-frequency-aware.md) (2026-09-02)
- [Tsinghua's Puro-2B Trains a Language Model for Less Than One GPU Rental](https://alphasignal.ai/news/tsinghua-s-puro-2b-trains-a-language-model-for-less-than-one-gpu-rental.md) (2026-09-01)
- [Tsinghua's SMELT Cuts AI Training Costs 18% by Looping Layers Twice](https://alphasignal.ai/news/tsinghua-s-smelt-cuts-ai-training-costs-18-by-looping-layers-twice.md) (2026-09-01)
- [CUHK's NoRA Fixes LoRA's Hidden Gradient Gap With one Line](https://alphasignal.ai/news/cuhk-s-nora-fixes-lora-s-hidden-gradient-gap-with-one-line.md) (2026-08-31)
- [Figure AI's Brett Adcock Bets Gigawatt NVIDIA Deal on Hark Handoff](https://alphasignal.ai/news/figure-ai-s-brett-adcock-bets-gigawatt-nvidia-deal-on-hark-handoff.md) (2026-08-27)
- [One Hidden Ratio Controls How Fast AI Models Actually Learn](https://alphasignal.ai/news/one-hidden-ratio-controls-how-fast-ai-models-actually-learn.md) (2026-08-25)
- [NovaSky's IsoExec Fixes the Hidden Math Bug Corrupting AI Training Runs](https://alphasignal.ai/news/novasky-s-isoexec-fixes-the-hidden-math-bug-corrupting-ai-training-runs.md) (2026-08-21)
- [Stanford's Marin Trains a 535B Open Model Anyone Can Watch Live](https://alphasignal.ai/news/stanford-s-marin-trains-a-535b-open-model-anyone-can-watch-live.md) (2026-08-21)
- [Kakao Trains a 155B MoE on 10T Tokens Without a Single Hyperparameter Sweep](https://alphasignal.ai/news/kakao-trains-a-155b-moe-on-10t-tokens-without-a-single-hyperparameter-sweep.md) (2026-08-20)
- [WhiteMatter Lets Any Transformer Layer Attend Across All Depths While Shrinking Memory](https://alphasignal.ai/news/whitematter-lets-any-transformer-layer-attend-across-all-depths-while-shrinking.md) (2026-08-19)