{"id":116010,"date":"2026-09-06T01:24:20","date_gmt":"2026-09-06T01:24:20","guid":{"rendered":"https:\/\/youzum.net\/adaption-labs-introduces-invent-a-dataset-training-data-generated-from-a-task-description-not-a-seed-corpus\/"},"modified":"2026-09-06T01:24:20","modified_gmt":"2026-09-06T01:24:20","slug":"adaption-labs-introduces-invent-a-dataset-training-data-generated-from-a-task-description-not-a-seed-corpus","status":"publish","type":"post","link":"https:\/\/youzum.net\/it\/adaption-labs-introduces-invent-a-dataset-training-data-generated-from-a-task-description-not-a-seed-corpus\/","title":{"rendered":"Adaption Labs Introduces \u2018Invent a Dataset\u2019: Training Data Generated From a Task Description, Not a Seed Corpus"},"content":{"rendered":"

This week, Adaption Labs<\/a> released Invent a Dataset<\/a>. The feature generates a structured, training-ready dataset from a description of the behavior you want a model to learn. You do not arrive with a seed corpus, a predefined schema, or a labeling guide. <\/p>\n

Is it deployable?<\/strong> Yes, with one caveat. Invent a Dataset is live now in the Adaption app and through the Python SDK and REST API<\/a>. Generated rows download as JSONL, JSON, CSV, or Parquet, so the artifact is a portable file you own and can train on anywhere. Generation itself runs on Adaption\u2019s hosted platform and consumes credits. No self-hosted generation path is documented.<\/p>\n

The problem being targeted<\/strong><\/h2>\n

Most dataset workflows begin with data that already exists. Teams then spend weeks labeling, filtering, and reshaping it to approximate the target task. Adaption\u2019s argument is that this caps model quality at how closely the available data matches the intended behavior. For proprietary and specialized tasks, the relevant signal usually sits in internal systems, unstructured text, or workflow logs. It rarely converts cleanly into a focused training set.<\/p>\n

The research team also draws a line against existing synthetic-data tooling. Those tools automate generation after a human has already defined the schema, task distribution, and generation strategy. Invent a Dataset starts one level earlier, at the behavior itself.<\/p>\n

How the API works<\/strong><\/h2>\n

The mechanics are documented and concrete. A single call to datasets.invent<\/code> creates the dataset and starts generation, returning immediately with status running<\/code>. You then poll datasets.get<\/code> until the status reads succeeded<\/code> or failed<\/code>, and download the rows.<\/p>\n

Domain codes are the primary control. You fetch current codes with datasets.invent_domains<\/code> rather than hardcoding them. You then pass values such as medical<\/code>, optionally narrowed by qualified subdomain codes such as medical.symptoms_diagnosis<\/code>. At least one domain or subdomain is required. Multiple domains contribute to the same run. A domain passed without subdomains draws from its full scope.<\/p>\n

Two output formats are supported. instruction_dataset<\/code> is the default and produces prompt-completion pairs for supervised fine-tuning. preference_pairs<\/code> produces chosen and rejected completions for preference-based training such as DPO.<\/p>\n

Three parameters matter for production use. estimate=True<\/code> prices the exact request and returns estimated versus available credits without creating or charging anything. prompt<\/code> accepts up to 10,000 characters to steer what the rows are actually about. idempotency_key<\/code> accepts up to 255 characters and makes network retries safe by returning the original dataset instead of launching a second run. Row counts are subject to a per-launch limit set by your plan.<\/p>\n

\n<\/div>\n

\n

Language and locale expansion<\/strong><\/h2>\n<\/p>

language_expansion<\/code> runs in two modes. translate<\/code> produces a new row variant for each target language. localize<\/code> produces a variant for each country and language pair, using locale-specific wording rather than direct translation. A sample_rate<\/code> between 0.01 and 1 controls what fraction of invented rows gets expanded, and credits are billed on the expanded output row count, not the original. Unsupported codes return a 400 with a sample of valid values.<\/p>\n

The zero-data loop<\/strong><\/h2>\n

Invent a Dataset is the first half of a loop. The dataset ID passes directly to autoscientist.create<\/code>, which co-optimizes the data and the training recipe against your objective. AutoScientist<\/a> launched in May 2026 and is the training-side counterpart to the Adaptive Data pillar.<\/p>\n

Adaption reports that AutoScientist beats training configured by its own research staff, by an average of 35%. Win rates moved from 48% to 64%. Those figures come from in-house domain-specialized evaluations across eight verticals. Dataset sizes ranged from 5,000 to 100,000 rows, on architectures offered for fine-tuning by Together AI<\/a>. <\/p>\n

Key Takeaways<\/strong><\/h2>\n