--- title: The Messy Reality of Scale: Synthetic Data and... description: Researchers from poolside (formerly Pulsar) share practical lessons from scaling their Laguna language models. On the data side, they describe using synthetic... canonical: https://daily.dev/posts/the-messy-reality-of-scale-synthetic-data-and-pre-training-marah-abdin-robert-mchardy-poolside-br07ki9ix twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: The Messy Reality of Scale: Synthetic Data and Pre-Training — Marah Abdin & Robert McHardy, poolside | daily.dev og:description: Researchers from poolside (formerly Pulsar) share practical lessons from scaling their Laguna language models. On the data side, they describe using synthetic... og:url: https://daily.dev/posts/the-messy-reality-of-scale-synthetic-data-and-pre-training-marah-abdin-robert-mchardy-poolside-br07ki9ix og:image: https://api.daily.dev/og/posts/BR07ki9IX.png og:image:alt: The Messy Reality of Scale: Synthetic Data and Pre-Training — Marah Abdin & Robert McHardy, poolside og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # The Messy Reality of Scale: Synthetic Data and Pre-Training — Marah Abdin & Robert McHardy, poolside **[AI Engineer](https://daily.dev/sources/aidotengineer)** · 17 min read · 0 upvotes · 0 comments ## Summary Researchers from poolside (formerly Pulsar) share practical lessons from scaling their Laguna language models. On the data side, they describe using synthetic data (13% of pre-training mix) not to replace organic data but to expose implicit structure, fill gaps, and address token repetition issues. They outline a modular synthetic data pipeline framework called Hive, covering rephrasing, multi-stage workflows, cross-domain porting, and multi-turn role pipelines. On the pre-training side, they detail real failure modes encountered at scale: silent GPU data corruption caught via model weight hash checks, and BF16 numerical precision failures in tensor-parallel accumulation near the LM head that caused training to stall — fixed by switching to FP32. They also reveal a race condition bug in DeepSpeed FP8 kernels causing ~0.5% silent gradient corruption, with a fix pending merge. Early benchmark results for the unreleased 118B parameter Laguna S model show strong coding and agentic performance compared to similarly-sized open-weight models. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [Scaling Laws for Mixture Pretraining Under Data Constraints](https://daily.dev/posts/scaling-laws-for-mixture-pretraining-under-data-constraints-c9l5eku29) · Apple Machine Learning Research · 0 upvotes · 0 comments - [「データ不足」の壁を越える：合成ペルソナが日本のAI開発を加速](https://daily.dev/posts/ai--xntmjjfpz) · Hugging Face · 1 upvotes · 0 comments - [Laguna XS.2 and M.1: A Deeper Dive — Poolside](https://daily.dev/posts/laguna-xs-2-and-m-1-a-deeper-dive-poolside-ko2qlww4d) · Hacker News · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#llm](https://daily.dev/tags/llm) [View this post on daily.dev](https://daily.dev/posts/the-messy-reality-of-scale-synthetic-data-and-pre-training-marah-abdin-robert-mchardy-poolside-br07ki9ix) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"The Messy Reality of Scale: Synthetic Data and Pre-Training — Marah Abdin & Robert McHardy, poolside","url":"https://daily.dev/posts/the-messy-reality-of-scale-synthetic-data-and-pre-training-marah-abdin-robert-mchardy-poolside-br07ki9ix","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/the-messy-reality-of-scale-synthetic-data-and-pre-training-marah-abdin-robert-mchardy-poolside-br07ki9ix"},"datePublished":"2026-07-26T01:05:26.642Z","dateModified":"2026-07-26T01:31:14.269Z","description":"Researchers from poolside (formerly Pulsar) share practical lessons from scaling their Laguna language models. On the data side, they describe using synthetic...","image":"https://i.ytimg.com/vi/KhYifX22yhE/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/KhYifX22yhE/sddefault.jpg","isAccessibleForFree":true,"articleSection":"AI Engineer","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"AI Engineer","logo":"https://media.daily.dev/image/upload/s--u5PucxNT--/f_auto/v1724338940/logos/aidotengineer","url":"https://daily.dev/sources/aidotengineer"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/the-messy-reality-of-scale-synthetic-data-and-pre-training-marah-abdin-robert-mchardy-poolside-br07ki9ix","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,llm","timeRequired":"PT17M","video":{"@type":"VideoObject","name":"The Messy Reality of Scale: Synthetic Data and Pre-Training — Marah Abdin & Robert McHardy, poolside","description":"Researchers from poolside (formerly Pulsar) share practical lessons from scaling their Laguna language models. On the data side, they describe using synthetic...","thumbnailUrl":"https://i.ytimg.com/vi/KhYifX22yhE/sddefault.jpg","uploadDate":"2026-07-26T01:05:26.642Z","duration":"PT17M","url":"https://api.daily.dev/r/BR07ki9IX","embedUrl":"https://www.youtube.com/embed/KhYifX22yhE"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AI Engineer","item":"https://daily.dev/sources/aidotengineer"},{"@type":"ListItem","position":3,"name":"The Messy Reality of Scale: Synthetic Data and Pre-Training — Marah Abdin & Robert McHardy, poolside"}]} ```