Top all-time · #24206
TileRT
Company blog · tilert.ai/blogaisystems
TileRT is the company blog at tilert.ai, writing about ai and systems. Since 2026, 3 different Hacker News readers have linked it 3 times, most often “Two Leaps to 1000 Tokens/s on a 1T-Parameter Model”. No feed was found, so read it on the site itself.
Too few sampled comments (1) to say how often HN recommends it.?
Why HN links it
HN readers link this TileRT blog for its inference optimization write-ups on large language models, such as the 1000 tokens-per-second post. The comment cites it for details of FP8 and FP4 quantization.
Summarised by Claude from a sample of 1 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
TileRT is a tile-level runtime engine that pushes the latency limits of large language models without compromising model size or quality.
Most-linked posts?
- Two Leaps to 1000 Tokens/s on a 1T-Parameter Model2 people · 2 mentions · first linked 2026-06-08
What Hacker News says?
I'm not a regular v2.5 user, so I don't really know. But given the TileRT team write-up says the entire network gets quantized to FP8 (and experts get quantized to FP4)[0], I'm assuming there's at least a modest drop in quality. [0] https://www.tilert.ai/blog/breaking-1000-tps.html
Comment linking /blog/breaking-1000-tps.html
on “Inference Optimization for MiMo v2.5: Pushing Hybrid SWA Efficiency to the Limit”Submitted TileRT Takes the Top Spot on AgentX with AMD Instinct MI355X GPUs as a story · 3 points
HN discussionSubmitted Two Leaps to 1000 Tokens/s on a 1T-Parameter Model as a story · 2 points
HN discussion