Skip to content

deepfabric

Model Serving & MLOps

A tool that generates synthetic training data for language models and agent evaluations by combining reasoning traces with tool‑calling patterns to produce high‑quality, domain‑specific datasets.

Track releases GitHub Website
Python Latest v4.12.0 · 4mo ago Security brief →

Features

  • Creates topic graphs (tree or graph) to ensure diverse yet relevant subtopic coverage
  • Generates training samples with constrained decoding, response validation, and real tool execution in isolated WebAssembly environments
  • Supports automatic upload of datasets to Hugging Face for use in frameworks like TRL, Unsloth, and Axolotl
  • Provides a built‑in evaluation engine to assess model performance on unseen tasks derived from training splits

Recent releases

View all 12 releases →
v4.12.0 Breaking risk
Breaking changes
  • Changed default mode for topic generation from unspecified to "graph"
Notable features
  • Added `inspect` and `prune` commands for topics
Full changelog

What's Changed

  • feat: add topic inspect and prune commands by @scp7 in https://github.com/always-further/deepfabric/pull/598
  • change: default mode for topic generation is now graph

Full Changelog: https://github.com/always-further/deepfabric/compare/v4.11.0...v4.12.0

View release on GitHub
v4.11.0 New feature
Notable features
  • UUID-based cycle generation with bug fixes and UX improvements
Full changelog

What's Changed

  • build(deps): bump astral-sh/setup-uv from 7.2.0 to 7.2.1 by @dependabot[bot] in https://github.com/always-further/deepfabric/pull/594
  • feat: UUID-based cycle generation with bug fixes and UX improvements by @scp7 in https://github.com/always-further/deepfabric/pull/595
  • refactor: compute tree topic IDs on-the-fly, remove persisted UUIDs by @scp7 in https://github.com/always-further/deepfabric/pull/596
  • test: integration test robustness — stubs, retries, async, diagnostics by @scp7 in https://github.com/always-further/deepfabric/pull/597

Full Changelog: https://github.com/always-further/deepfabric/compare/v4.10.1...v4.11.0

View release on GitHub
v4.10.0 Breaking risk
Breaking changes
  • Removed multi turn agent mode
Notable features
  • Auto and percentage-based num_samples configuration
  • Checkpoint v2 with XDG paths, graceful stop, and UX enhancements
Full changelog

What's Changed

  • docs(training): update documentation regarding training extra by @ognis1205 in https://github.com/always-further/deepfabric/pull/581
  • Remove multi turn agent mode by @lukehinds in https://github.com/always-further/deepfabric/pull/584
  • build(deps): bump transformers from 5.0.0rc2 to 5.0.0rc3 by @dependabot[bot] in https://github.com/always-further/deepfabric/pull/582
  • chore: add GitHub issue and PR templates by @ognis1205 in https://github.com/always-further/deepfabric/pull/564
  • feat: add auto and percentage-based num_samples configuration by @scp7 in https://github.com/always-further/deepfabric/pull/585
  • Feature checkpointing v2 by @scp7 in https://github.com/always-further/deepfabric/pull/589
  • build(deps): bump actions/setup-python from 6.1.0 to 6.2.0 by @dependabot[bot] in https://github.com/always-further/deepfabric/pull/587
  • build(deps): bump actions/checkout from 6.0.1 to 6.0.2 by @dependabot[bot] in https://github.com/always-further/deepfabric/pull/588
  • feat: checkpoint improvements - XDG paths, graceful stop, and UX enhancements by @scp7 in https://github.com/always-further/deepfabric/pull/590

Full Changelog: https://github.com/always-further/deepfabric/compare/v4.9.0...v4.10.0

View release on GitHub
v4.8.3 Breaking risk
Breaking changes
  • Removed the mcp-adversarial-suite component.
Notable features
  • Added DeepFabric dataset generation documentation.
  • Introduced prompt_style configuration option.
Full changelog

What's Changed

  • Remove mcp-adversarial-suite by @lukehinds in https://github.com/always-further/deepfabric/pull/560
  • build(deps): bump astral-sh/setup-uv from 7.1.6 to 7.2.0 by @dependabot[bot] in https://github.com/always-further/deepfabric/pull/561
  • docs: add CODE_OF_CONDUCT.md by @ognis1205 in https://github.com/always-further/deepfabric/pull/563
  • chore: migrate PostHog to EU host by @scp7 in https://github.com/always-further/deepfabric/pull/540
  • docs: add CONTRIBUTING.md by @ognis1205 in https://github.com/always-further/deepfabric/pull/577
  • Add DeepFabric dataset generation docs and prompt_style config by @lukehinds in https://github.com/always-further/deepfabric/pull/578
  • Introduce prompt_style to configs by @lukehinds in https://github.com/always-further/deepfabric/pull/579

Full Changelog: https://github.com/always-further/deepfabric/compare/v4.8.2...v4.8.3

View release on GitHub
v4.7.1 Bug fix

Fixed HuggingFace Dataset compatibility in native Dataset.map().

Full changelog

What's Changed

  • fix: add HuggingFace Dataset compatibility to native Dataset.map() by @lukehinds in https://github.com/always-further/deepfabric/pull/546

Full Changelog: https://github.com/always-further/deepfabric/compare/v4.7.0...v4.7.1

View release on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

About

Stars
872
Forks
83
Languages
Python Jupyter Notebook Rust
View on GitHub Homepage

Install & Platforms

Install via
pip

Community & Support

Discord Community

Similar tools

Modafinil
Velociraptor
Shdoc

Beta — feedback welcome: [email protected]