Toridion/lindisfarne-m1
收藏资源简介：
Lindisfarne M1是一个完全合成的、包含一百万条NHS（英国国家医疗服务体系）风格健康记录的语料库，由Toridion TQNN DMM合成数据管道生成。记录基于真实的NHS临床数据结构，包括SNOMED CT、HL7 v2.5、FHIR R4和GHX变体，但不包含任何真实患者信息；所有姓名、NHS号码、国民保险号码、地址、邮政编码、诊断、药物和临床医生信息均为程序生成。数据集分为三个层次以满足不同用例：Full层（20-35个字段，丰富嵌套记录，包含所有格式、实验室、过敏、警报和程序）、Slim层（10-15个字段，减少嵌套，最多2个叙述字符串，摄取速度更快）和Superslim层（10个扁平字段，扁平JSON，无数组，摄取速度最快）。每个层次包含1,000,000条JSONL格式记录，每文件10,000条记录（每层次100个文件）。记录格式分布约为34% HL7 v2.5、34% FHIR R4、28% SNOMED CT和4% GHX。数据集旨在用于搜索基准测试、DMM摄取测试、开发者验证（如ETL管道、解析器、索引器）和ML/AI研究（如预训练、微调或评估），无需担心患者隐私风险。数据集生成使用合成生成器，确保无真实数据，包括随机生成的姓名、NHS号码、地址等，并支持通过种子可重复输出。数据集完全安全，不包含任何真实患者信息，适合公开分发和商业使用。
Lindisfarne M1 is a fully synthetic corpus of one million NHS-style health records generated by the Toridion TQNN DMM synthetic data pipeline. Records are modelled on real NHS clinical data structures — SNOMED CT, HL7 v2.5, FHIR R4, and a GHX variant — but contain no real patient information of any kind. Every name, NHS number, NI number, address, postcode, diagnosis, medication, and clinician is procedurally generated. The dataset is published in three tiers: Full (20–35 fields, rich nested records, all formats, labs, allergies, alerts, procedures), Slim (10–15 fields, reduced nesting, max 2 narrative strings, faster ingest), and Superslim (10 flat fields, flat JSON, no arrays, maximum ingest speed). All three tiers contain exactly 1,000,000 records in JSONL format, split into chunks of 10,000 records per file (100 files per tier). Record formats are distributed as approximately 34% HL7 v2.5, 34% FHIR R4, 28% SNOMED CT, and 4% GHX. Intended uses include search benchmarking, DMM ingest testing, developer validation (e.g., ETL pipelines, parsers, indexers), and ML/AI research (e.g., pre-training, fine-tuning, or evaluation) without patient privacy risk. Data generation uses a synthetic generator with no real data, including randomly generated names, NHS numbers, addresses, etc., and supports reproducible output via seed. The dataset is entirely safe, containing no real patient information, suitable for public distribution and commercial use.
