131K-token patient record read in 61s, 11 minutes of reasoning on 2x DGX Spark
maniac518 · reddit · 2026-11-19
Developer SudoingX fed AntLing's Ling-3.0-flash-Sante a synthetic patient's full record — 334 visits across 71 years, 130,993 tokens in one prompt — on two DGX Sparks with official FP8 weights. Reading took 61 seconds (2,148 tok/s); the model then reasoned for 30,865 tokens, about 11 minutes.
Scored against an answer key: 7/7 allergies, 8/10 active conditions, 2/7 dental codes, 2/5 medications; no stopped items leaked onto active lists, and it caught a medication started in 1954 and still running 67 years later. The author notes it's a research model, not a medical device. The hard part: dense start/stop/restart entries where 'current' depends on getting every date right.
More from Infra
- NVIDIA to demo Cosmos Data Factory pipeline for training-ready synthetic datasets with Voxel51 and Encord — NVIDIAAI · 2026-11-19
- Analyst: memory has shifted from commodity to value, explaining sustainably higher margins — BenBajarin · 2026-11-19
- Microsoft open-sources Aion 2.3B on-device SLM, shipping as the Windows built-in model — jacek2023 · 2026-11-19
- Local model reads a 131K-token patient history in 61s, then reasons for 11 minutes — maniac518 · 2026-11-19
- Voters rejected 46 of 47 local data center ballot measures, analysis finds — WillRinehart · 2026-11-19
- McKinsey roadmap says quantum could create $2.7T in value by 2035 — ChuckDBrooks · 2026-11-19