# bespokelabsai/curator

Synthetic data curation for post-training and structured data extraction

Repository: https://github.com/bespokelabsai/curator
Canonical: https://ross.abutalabs.com/products/bespokelabsai-curator
Homepage: https://docs.bespokelabs.ai/bespoke-curator
Language: Python
License: Apache-2.0
License Family: permissive
Topics: synthetic-data, agents, llm, prompt, python, synthetic-dataset-generation, deep-learning, fine-tuning, instruction-tuning, machine-learning, natural-language-processing
Last push: 2026-08-07T07:54:05+00:00

## Health v2 (maintenance only)
Score: 79/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 96, release rhythm 75, longevity 48
- inputs: {"age_days": 675, "days_push": 26, "days_rel": 171, "gap_med": 7, "n_releases_24m": 26}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1719, forks 145 (observed 2026-08-28T04:05:26.690622+00:00)

## What it is
Bespoke Curator is a Python library for bulk LLM inference and scalable synthetic data curation for post-training and structured data extraction. It provides a declarative prompt/parse interface over datasets, with caching, fault recovery, structured outputs via Pydantic, and integrations with LiteLLM, vLLM, batch APIs, and fine-tuning backends like Tinker and Fireworks AI.

## Use cases
- generate synthetic instruction-tuning datasets with llms
- bulk run prompts over a dataset in parallel
- extract structured data from llm responses with pydantic schemas
- create reasoning datasets for fine-tuning like openthoughts
- resume large llm data generation runs after failures
- fine-tune a lora from curated data via tinker or fireworks
- visualize and inspect llm-generated datasets while generating

## When to choose
- you need to generate large-scale synthetic training data with llms reliably
- you want structured outputs and caching/fault recovery built into a data pipeline
- you are building post-training datasets (sft, reasoning, agents) for fine-tuning

## When to avoid
- you need a single one-off llm call rather than dataset-scale generation
- you want a no-code gui tool instead of a python library
- you are not comfortable sending data through hosted llm apis or the hosted viewer

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, data-generation, etl, rag, llm-training, sdk
- domain: machine-learning, large-language-models, developer-tools
- platform: python, cross-platform
- tags: synthetic-data, fine-tuning, structured-output, batch-inference, data-curation, instruction-tuning, litellm, vllm, data-engineering, natural-language-processing

## Member repositories
- bespokelabsai/curator (main) score 79

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:26.690622+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:34:06.679988+00:00, confidence not recorded.
  - readme: https://github.com/bespokelabsai/curator (fetched 2026-08-28T04:05:26.690622+00:00, sha 94d910ba507a)
  - homepage: https://docs.bespokelabs.ai/bespoke-curator (fetched 2026-08-29T11:10:04.077729+00:00, sha c7eb63cf550b)
  - site_page: https://docs.bespokelabs.ai/bespoke-curator/getting-started/structured-output (fetched 2026-08-29T11:10:04.095137+00:00, sha af71c629c6ce)
  - site_page: https://docs.bespokelabs.ai/bespoke-curator/getting-started/quick-tour (fetched 2026-08-29T11:10:04.087054+00:00, sha 90b75f8e091c)
  - site_page: https://docs.bespokelabs.ai/bespoke-curator/getting-started/key-concepts (fetched 2026-08-29T11:10:04.089444+00:00, sha 02f579f98a73)
  - site_page: https://docs.bespokelabs.ai/bespoke-curator/getting-started/visualize-your-dataset-with-the-bespoke-curator-viewer (fetched 2026-08-29T11:10:04.091521+00:00, sha d975e459e0e9)
  - site_page: https://docs.bespokelabs.ai/bespoke-curator/getting-started/automatic-recovery-and-caching (fetched 2026-08-29T11:10:04.093315+00:00, sha a41191615d71)
- Data as of 2026-08-30T08:39:29.467469+00:00.