Xiaomi Mimo 2. 6 live post-training dashboard

Xiaomi Mimo 2. 6 live post-training dashboard. A Postgres database should know everything about the stuff that lives in its tables, no?

What happened

Image caption, Ukraine's post office provides a vital lifeline for people near the front, delivering pensions and medicineOne customer, Nadiia, said she was grateful: "It's a long way for us to travel to Zaporizhzhia. Qwen produces a candidate strategy per rollout and sends it off to Postgres for measurement against its own default plan. What follows is a breakdown of an experiment I ran to explore the question: can a small, open-weights model be post-trained via supervised fine-tuning (SFT) and agentic reinforcement learning (RL) to produce Postgres query plans that beat Postgres’s default plans?

7% latency reduction across 113 join-heavy queries from a 4B model initially unable to produce a query plan for 99 of them Constructing a Postgres measurement rig that minimizes Linux page cache contention noise across concurrent containers Designing a custom GRPO variant for scoring RL rollouts in an inherently noisy environment Splitting RL across two machines: vLLM and the trainer on a rented 2x H100 node and four Postgres containers running on my desk Running off-policy distillation across half a thousand GPT-6 Astra agent trajectories Let’s start from the beginning.

The path Postgres took to get this data for us is not a foregone conclusion, and it has everything to do with what we call selective predicates (i. e. the filtering conditions in a WHERE clause). It’s also worth noting that Postgres doesn’t evaluate all of these plans. Figure caption, Watch: Ukraine's post office innovation helps isolated communitiesByJames LandaleReporting fromZaporizhzhia oblastPublished15 September 2026Every morning Igor Smelyansky looks at a vast map of Ukraine on his office wall and decides where he should put his people in harm's way.

The wider picture

His aides provide him with live intelligence about Russian attacks and he deploys his teams, balancing their potential operational success against risk to life or limb. He is the head of Ukraine's state postal service, Ukrposhta. And every day he must decide how far to send his postmen and women into the deadly "kill zone" along the front lines where the threat of drone attack is ever present. And it delivers vital food and medicine to people on the margins of society.

That has made the service a target for Russia – which in recent months has stepped up not only its attacks on postmen and women but also sorting offices and logistical hubs. If a post van turns up unannounced, a crowd can form, creating a target for an opportunistic Russian drone operator. "Smelyansky says Russia is deliberately targeting his staff because of the vital role they play shoring up Ukraine's resilience, helping people stay alive in their homes through the fighting and bitter winters.

Since 2022, hundreds of Ukrposhta branches and vehicles have been hit, and more than 50 staff have lost their lives, either on the frontline or on their rounds, according to Ukrposhta.

What has been reported

Image source, Getty ImagesImage caption, A file photo of a Ukrposhta post office damaged during strikes in Kyiv in AprilWe travelled with Kateryna Marynych, a 38-year-old postmistress, near Zaporizhzhia – a city not too far from the southern frontlines. She ran the post office in the village of Lukasheve for two years until it was destroyed in March by a Russian drone, targeting soldiers working nearby. "But Mykola, 64, pointed to another service the post office provides to a largely elderly population.

"Image caption, Kateryna Marynych has lost her home and her husband in this war – now, she's also lost her post officeThe attacks on Ukrposhta form part of Russia's wider escalation of its war on Ukraine, stepping up fast drone attacks on civilian infrastructure including business infrastructure, energy facilities, transport networks and border posts. While she sorts the cash, he works from the back of the van, taking and receiving parcels, delivering food and water.

"Ukraine's defence comes in many forms – and it's not always delivered at the barrel of a gun. Skip to article A reinforcement-learning policy update Four RL rollouts commence for a single query. Scalar rewards are assigned to each rollout, which flow backwards to update Qwen's weights.

What happens next

Despite an enormous body of research spanning a decade since their original exploration, they found that query optimizers continue to leave much to be desired. I was surprised when I first learned about this. In fact, one particular task a query optimizer needs to do, join ordering, is known to be NP-hard. What’s not as hard is verifying whether a query plan an optimizer picks is good or not. Put simply, a good query optimizer produces plans that run fast, and a bad one produces slow plans.

Language models are particularly good at learning how to do tasks with easily verifiable outputs. Because there’s a single axis to optimize for—execution time of a query—the problem beautifully reduces to reinforcing the behaviors that guide a model to produce faster query plans.

The report has been compiled by The Daily Waves using information reported across mimo.xiaomi.com, rohanbansal.com, bbc.co.uk. Details are presented according to the information available at the time of publication and may change as authorities, organisers or other relevant parties provide updates.