Blogsby GlyphHow Novita Sandbox Powers Miles' Harbor and OpenEnv RL Training Recipes
Novita Sandbox is now a Miles sandbox provider. Run the Harbor and OpenEnv RL recipes with copy-paste commands, configuration variables, and defaults.
Machine learning
Training, recommendation, ranking, feature stores and ML platforms. 812 posts on file, most from Runpod, CoreWeave, AWS.
Fri Nov 67 posts
Zero Mismatch, Simple Async RL: REINFORCE Zero, DeepSeek-V4.1 and Kimi K3.1
Asynchronous RL trains on stale weights. Align rollout and training kernels bit for bit, and unclipped importance sampling corrects that lag. Plus zero-mismatch RL for DeepSeek-V4.1-Flash, Kimi K3.1 and AMD MI355X.
+1more from vLLM Blog that day
ToyBlocks: A controlled world for robot learning and evaluation
ToyBlocks is an open simulation framework for studying robot learning through automated task generation, training demonstrations, and evaluation in a controlled tabletop world.
Safe and Efficient Multi-Epoch Embedding Training
Authors: Shruthi Hariharan | Machine Learning Engineer II; Konish Dutta | Machine Learning Engineer II; Abhinav Naikawadi | Sr. Machine Learning Engineer; Zhihua Zhang | Machine Learning Engineer II; Rakshita Nagalla | Sr.
Accelerating Reinforcement Learning Weight Refit with the Reshard Collective
During reinforcement learning (RL) post-training, training trajectories are generated on the fly by running inference with the current policy model. For...
SageMaker Inference now supports Multi-Instance GPU to maximize GPU utilization for model endpoints
Amazon SageMaker Inference now supports GPU partitioning with NVIDIA Multi-Instance GPU (MIG), which divides one GPU into hardware-isolated partitions so you can run a different model in each or take a right-sized fraction of a newer...
Building Runpod templates: Your first custom template
Step-by-step guide to building a custom PyTorch Docker image, pushing it to Docker Hub, and saving it as a reusable Runpod Pod template with a GPU test.
Thu Nov 53 posts
Searchable now tracks Meta Muse: see whether your brand makes the shortlist
Searchable now tracks Meta Muse. See whether your brand appears in AI shopping recommendations, how it compares with competitors and where to improve.
+1more from Searchable Blog that day
Why SUNK Redefines the AI Research Cluster for Production-Grade Training
Built for long-running, thousand-GPU jobs, SUNK (Slurm on Kubernetes) delivers predictable and performant AI training through topology-aware scheduling and continuous health management.
Choosing How ML Models Move to Production on Snowflake: Code, Model, or Hybrid Promotion
Data access, training cost, CI/CD readiness, and business needs decide how your models reach production, and shape your environments, your pipelines, and your risk.
Wed Nov 42 posts
Enterprise ML Architecture: Making Every Boundary Earn Its Place
Seven Workload Tests for Better Architecture Decisions Enterprise ML architectures carry history. Many were built when the data platform handled storage, transformation and analytics, while a separate ML platform supplied notebooks...
NVIDIA CUDA 13.5 Brings NVIDIA Vera Rubin GA Support
As artificial intelligence and accelerated computing continue to push the boundaries of what's possible, the latest NVIDIA CUDA release arrives to unlock...
Tue Nov 33 posts
F3 Lightning: How Meta Accelerates Ads ML Features to Production at Scale
Facebook Feature Framework (F3) is Meta’s framework for authoring, compiling, and serving the machine learning features underlying ads ranking models, where a single request can require more than 20,000 features.
Frontier quality at 1% of the cost: a production benchmark of LLM classifiers
We benchmarked a fine-tuned E2B model against frontier APIs on a real production classification task. It matched their quality in under 100ms, at about a hundredth of the cost.
Fri Oct 302 posts
Fair use benefits everyone
OpenAI shares excerpts from court briefs by creators, academics, accessibility advocates, and legal experts, among others, on how fair use supports AI training and innovation.
+1more from OpenAI News that day
Wealth Management Client Onboarding Needs More Than AI
Wealth Management client onboarding can stall even when every system has done its job. The document is collected...
Thu Oct 296 posts
Serverless GPU inference vs. GPU Pods: cost benchmarks
We benchmarked one A40 as a Pod and as a Serverless worker. A busy worker cost about 2.3x per token, but Serverless wins below 29 billed minutes an hour.
Nutanix AHV Earns NVIDIA Certification for AI Workloads
Nutanix AHV earns certification under NVIDIA's Certified Hypervisors program, delivering validated GPU-accelerated AI performance on Nutanix Cloud Platform.
What to Do About Dark Data Before Your Next AI Pilot
What to Do About Dark Data Before Your Next AI Pilot by Everpure Blog See how to prepare dark data for AI with a four-step readiness check: discover, classify, contextualize, and vectorize only the governed data your next pilot needs.
Play the future of mobile gaming: Neural Dawn now available to download
Co-developed by Arm and Sumo, Neural Dawn is now available to download and play on the latest mobile devices powered by the Arm Mali G2-Ultra NX GPU, bringing cinematic lighting, ray tracing, and neural graphics directly to the hands of...
How we’re training a Figma-fluent editing model
We’ve increasingly seen that specialized post-training can compete with frontier models. Here, we’re sharing the research behind how we’re translating that to design editing workflows.
Unlocking Tensor Core Performance for Machine Learning Interatomic Potentials
Atomistic simulations help researchers understand how molecules and materials behave, from chemical reactions and phase transitions to mechanical failure. These...
Wed Oct 283 posts
Why Banking AI Succeeds or Fails at the Data Foundation
Why Banking AI Succeeds or Fails at the Data Foundation by Everpure Blog Learn why banking AI pilots stall and how trusted, governed data and resilient infrastructure can help banks move those projects into production.
Distributed training with Ray Train DLCs and EFA on EKS
The Ray Train Deep Learning Container ships the full CUDA-through-EFA stack, so you build no image. This post walks through a reproducible sample that fine-tunes Qwen2.5-1.5B with PyTorch FSDP across eight GPUs on two EFA-enabled...
Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design
This post describes the next generation of Meta Andromeda, Meta’s personalized ads retrieval engine, and how co-designing the model, system, and GPU hardware unlocks a new efficiency frontier for ads retrieval.
Tue Oct 272 posts
Vessel-specific voyage optimization with machine learning: Manta on AWS
Manta Marine Technologies closes the gap between seemingly identical ships by training machine learning models on each vessel's own operational history.
Auto repair for Together GPU Clusters: from GPU fault to healthy node
Together GPU Clusters auto repair detects GPU, host and fabric faults, picks the lightest fix, and lets you decide per fault and per Xid code which repairs run automatically. Here is how it works.
Mon Oct 262 posts
Nearly 3 in 4 enterprises hear customer complaints on AI personalization
Most enterprises get measurable results from AI personalization, but many get customer complaints. See the data on what breaks and how teams find out.
Fleet-wide computer vision lifecycle at the edge with SageMaker SDK v3
Fleet-wide computer vision needs a model per camera, not one model for the whole fleet. This post presents a working Amazon SageMaker SDK v3 reference architecture that trains, quality-gates, compiles per-camera models with SageMaker Neo...
Fri Oct 235 posts
Security Disclosure Recommendation: Recover Failing Deploys
Commits pushed while Railway paused deployments on October 21 were not deployed. Redeploy the latest commit, reconnect GitHub, and recreate CLI and CI credentials to get your services deploying again.
+1more from Railway Blog that day
Readiness Is Not Curiosity: Why Your Agency’s AI Program Is Waiting on People, Not Models
Readiness Is Not Curiosity: Why Your Agency’s AI Program Is Waiting on People, Not Models by Everpure Blog The pilot worked, but the agency didn't change.
Post-Training at the Frontier
RL Post-Training at the Frontier: How CoreWeave, You.com and NVIDIA Customized Nemotron’s Agentic Search & Browsing Capabilities with RL Rollouts
Multi-Turn Reinforcement Learning on Amazon SageMaker AI: A hands-on guide
Multi-Turn Reinforcement Learning (MTRL) on Amazon SageMaker AI trains models on the outcome of a whole multi-step task rather than labeled steps.
Securing our frontier model training environment
OpenAI explains how we strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.
Thu Oct 225 posts
Collibra + Trail ML: Bringing AI governance and automation together
Most organizations have moved past asking whether to use AI. The harder question now is whether they can prove the AI they run is doing what it should. That gap is why we are writing this together.
Introducing our Training API
Spindle is an open-source API for training and fine-tuning open models, deploy a server with a single command and run any Tinker-compatible training code against it.
Write your own training loop, without managing GPUs, with Amazon SageMaker training session APIs
Amazon SageMaker AI training session APIs let you write your own LLM training loop (SFT, RL, or custom objectives) in a Python script while the service runs the GPUs.
+1more from AWS Machine Learning Blog that day
Security Disclosure Recommendation: Check for Unexpected OpenRouter Usage
Learn how to Security Disclosure Recommendation: Check for Unexpected OpenRouter Usage
+10more from Railway Blog that day
- Security Disclosure Recommendation: Check for Unexpected Grok Usage
- Security Disclosure Recommendation: Redeploy Buckets and Reconnect GitHub After the Credential Reset
- Security Disclosure Recommendation: Check for Unexpected Anthropic Usage
- Security Disclosure Recommendation: Check for Unexpected OpenAI Usage
- Security Disclosure Recommendation: Check for Unexpected Usage in Stripe
- Security Disclosure Recommendation: Check for Anomalous GitHub Access
- Security Disclosure Recommendation: Rotate MongoDB Database Credentials
- Security Disclosure Recommendation: Rotate Redis Database Credentials
- Security Disclosure Recommendation: Check for Anomalous AWS Access
- Security Disclosure Recommendation: Check for Anomalous GCP Access
Rebuilding our PyTorch templates
Runpod's PyTorch templates now span 2.6 to 2.13 with multiple CUDA and Ubuntu versions, plus a CUDA dropdown in the Pod deploy flow.
+6more from Runpod Blog that day
- Get Better DreamBooth Results Using Offset Noise
- Run SAM 2 on a Cloud GPU with Runpod (Step-by-Step Guide)
- Run the Flux Image Generator on Runpod (Full Setup Guide)
- How to Connect Google Colab to Runpod
- How to Run SAM 2 on a Cloud GPU with Runpod
- Run Flux Image Generator in ComfyUI on Runpod (Step-by-Step Guide)