5 results for large scale là gì

github.com
Show HN: K7d – Fork live Kubernetes clusters in GRPO-train AI on infrahttps://github.com/katakate/k7dHey HN, Gary here.Today I want to present k7d which is an Apache 2.0, tight Rust VMM + shim enabling something not possible before: fast forking of live running virtualized multi-node k8s clusters with surviving of in-flight connections.A 3-VM nodes K8s cluster gets forked in 105ms, and a 50x fork of a 3-VM cluster in 4.1s on a 64GB RAM box.I have two goals here:1) enable large scale GRPO/RL training of AI on Kubernetes infra, which IMO is a great playground for reasoning training, besides training a capability that's actually useful. And this requires not only fast episode reset (as you need tens of thousand of multi-turn runs during RL post-training) but also greatly benefits from fast forking so you can do parallel branch exploration, rollback, pruning during RL training. Faithful forks also give you byte-identical starts for the G of GRPO, which gives variance reduction across the group.2) enable
Aug 18, 2026 2:38 PM
news.ycombinator.com
Ask HN: Model access depends on citizenship. What should Non-US founders do?https://news.ycombinator.com/item?id=48692825LLMs are at this point a necessity to develop large scale systems - we couldn't do what we do without them. Our team is worried by the developments of Mythos/Fable getting restricted and the news about GPT 5.6 access being gated by the government customer-by-customer.The part that complicates the obvious answers: the Anthropic directive keyed on the nationality of the person accessing the model, not where the company is incorporated; even their own foreign-national employees lost access. So "just flip to a Delaware C-corp" doesn't clearly solve it for a team like ours.This is all new and we don't think anyone has real answers yet. We want to hear what others think might work. A few directions we've floated, none of which we're sure about:- Hoping that at least one frontier model provider remains accessible - An open-weight/non-US floor (GLM/DeepSeek/Qwen) - Hoping for frontier access via Bedrock/Vertex/Azure rather than first-party APIsOur concern is that the regional capability differ
Jun 26, 2026 10:29 PM
news.ycombinator.com
A 30 Year OG Application Developer Availablehttps://news.ycombinator.com/item?id=48668766https://www.youtube.com/watch?v=DACtpW9Q-hcThat link is the #1 Architectural Interior Design software used by the top firms in the world. What you are looking at is the result of pure architectural discipline. To handle massive global budgets, complex subcontractor workflows, and real-time synchronization with massive enterprise CAD suites, I engineered a closed-loop, self-aware data object model. The data objects carry their own application logic and database schemas. It is a zero-exception, self-contained architecture that eliminates the need for lazy, nested error handling. It runs perfectly because it mathematically cannot run corruptly. It has a bidirectional interface to Revit, the same software Boeing used to design their jets.I bring 30+ years of deep engineering experience across both Microsoft and Linux stacks. I am a SQL/relational database expert, a masterful architect of rugged backend APIs, and highly experienced in deep enterprise integrations.I am out of the large-scale
Jun 25, 2026 4:09 AM
spacedaily.com
Smartphone kit offers low cost on site radiation dose checkshttps://www.spacedaily.com/reports/Smartphone_kit_offers_low_cost_on_site_radiation_dose_checks_999.htmlLos Angeles CA (SPX) Jan 29, 2026 Prompt, individual dose assessment is a critical part of protecting people after large scale nuclear or radiological incidents, but traditional dosimetry relies on expensive instruments or laboratory processing that can delay decisions on medical care. Researchers at Hiroshima University have now demonstrated a portable, low cost system that pairs a small piece of radiochromic film with a foldab
Feb 3, 2026 11:07 AM