Watch a 535B parameter LLM with 23B active parameters being trained live
iskander · x · 2026-08-21
A live stream showcases the training of a massive 535B parameter LLM (with 23B active parameters). The link provides real-time access to the training process, likely utilizing a sparse activation architecture.
More from Infra
- IBM launches Docling, a free Python library to convert any document to data — mdancho84 · 2026-08-21
- Starcloud raises $250M at $2.3B valuation to build AI compute in orbit — ryanbed · 2026-08-21
- Netlify Launches 'Source': Git Power Without GitHub Accounts — thisiskp_ · 2026-08-21
- BF16 Checkpoint Gotcha: +1 Followed by -1 Isn't Always a Round Trip — ReinforcedKnowledge · 2026-08-21
- Leak: Google is the mystery hyperscaler behind Modine's $4B deal — citrini · 2026-08-21
- AMD reports 4x AI rack-scale efficiency gain since 2024, ahead of target — ryanshrout · 2026-08-21