Skip to content

Releases

v2.7.0

Choose a tag to compare

@github-actions github-actions released this 13 Nov 10:06
Immutable release. Only release title and notes can be modified.

What's Changed

  • Checkpoint eval defaults: default task is now evals/kmmlu_redux.py across SFT, DPO, OPD, SDFT and AsyncGRPO (GRPO uses inspect_evals/math); DEFAULT_EVAL_TASKS and docs updated to match
  • AsyncGRPO defaults restructured with section headers; eval still runs on eval_devices: "2" since the rollout server sits on GPU 1
  • OPD KMMLU recipe: constant LR with warmup, 5 epochs
  • New ZeRO-2 and ZeRO-3 CPU offload accelerate configs
  • Removed redundant qwen3-0.6B-sft config and configs/DPO/qwen3-0.6B-dpo.yaml

Each release carries a CycloneDX SBOM (sbom-2.7.0.cyclonedx.json + .sha256) generated from uv.lock.

Full Changelog: v2.6.0...v2.7.0

v2.6.0

Choose a tag to compare

@github-actions github-actions released this 07 Nov 07:29
Immutable release. Only release title and notes can be modified.

dna-factory 2.6.0 renames on-policy distillation to OPD (On-Policy Distillation) in the code, configs and docs. Training behavior is unchanged.

What's Changed

  • distill.py is now opd.py, with configs in configs/OPD/ (_defaults-OPD.yaml, *-opd*.yaml), DnotitiaOPDTrainer in dna_factory/dnotitia_opd_trainer.py, docs/opd.md and utils/measure-opd-kl.py. (#45)
  • output_dir: auto now tags OPD runs OPD instead of DISTILL.
  • The YAML section header and the startup log header are now DNOTITIA TRAINER ARGUMENTS for every entry point, and the unused TrainingSpec.trainer_type field is removed.

Full Changelog: v2.5.2...v2.6.0

v2.5.2

Choose a tag to compare

@github-actions github-actions released this 06 Nov 16:20
Immutable release. Only release title and notes can be modified.

Docs-only patch release on the 2.5 line. The code is the same as v2.5.1.

Docs

  • The README intro now covers all five trainers, including SDFT (Self-Distillation Fine-Tuning).
  • On-Policy Distillation (distill.py) is now called OPD in the README and docs/. Paper and blog titles keep their original names.
  • The README news bolds only the acronyms for GRPO and DPO.
  • Added NIA to the README acknowledgments.

Supply chain

  • sbom-2.5.2.cyclonedx.json (CycloneDX) and its .sha256 checksum are attached. Verify the download, then scan it for known vulnerabilities:

    sha256sum -c sbom-2.5.2.cyclonedx.json.sha256
    grype sbom:sbom-2.5.2.cyclonedx.json

Full Changelog: v2.5.1...v2.5.2

v2.5.1

Choose a tag to compare

@github-actions github-actions released this 06 Nov 10:25
Immutable release. Only release title and notes can be modified.

Patch release on the 2.5 line. It is the first release that ships a CycloneDX SBOM. The code is the same as v2.5.0; see that release for the new features (SDFT, the unsloth backend, Qwen3.8-Flash-Next LoRA SFT, KMMLU recipes, and the TRL 1.17.1 / transformers 5.22.1 bump).

Supply chain

  • SBOM on every release: sbom-2.5.1.cyclonedx.json (CycloneDX 1.7) lists all 300 packages pinned in uv.lock, including the unsloth extra. Its .sha256 checksum is attached alongside it. Verify the download, then scan it for known vulnerabilities:

    sha256sum -c sbom-2.5.1.cyclonedx.json.sha256
    grype sbom:sbom-2.5.1.cyclonedx.json
  • Releases in this repo are immutable, so the SBOM is now attached while the release is still a draft. Pushing a version tag creates the draft. The steps are in docs/release.md. v2.5.0 was published before this change and has no SBOM.

Docs

  • Shortened the SDFT comment in the README quick start.

Full Changelog: v2.5.0...v2.5.1

v2.5.0

Choose a tag to compare

@likejazz likejazz released this 06 Nov 10:21
Immutable release. Only release title and notes can be modified.

dna-factory 2.5.0 adds two new training methods (SDFT and an opt-in unsloth backend), Qwen3.8-Flash-Next LoRA SFT, and KMMLU recipes for GRPO and on-policy distillation. It also bumps TRL to 1.17.1 and transformers to 5.22.1.

New Features

  • SDFT (Self-Distillation Fine-Tuning): new sdft.py entry point. The teacher is the model itself, shown a per-row privileged_context that the student does not see. It trains on GSM8K demonstrations by default. See docs/sdft.md. (#43, @cm8908)
  • unsloth backend: opt-in with use_unsloth: true for SFT, DPO, GRPO and distillation (uv sync --extra unsloth). Single GPU only. See docs/unsloth.md. (#34, @cm8908)
  • Qwen3.8-Flash-Next LoRA SFT, plus hardening of its training path. See docs/qwen38-flash-next.md. (#40, @cm8908)
  • Faster qsa indexer: vectorized across the batch and its padding modes. (#42, @cm8908)
  • Thinking-model GRPO rewards that close accuracy leaks (dna_factory/rewards/reasoning.py).
  • KMMLU recipes: GRPO with multiple-choice rewards, and on-policy distillation of Qwen3-4B from Qwen3-32B.

Changes

  • GRPO defaults: halved default completion and eval lengths, vllm_max_model_length capped at 12480, and vllm_importance_sampling_correction set explicitly.
  • Distill defaults: vllm_max_model_length capped.
  • GRPO config variants (no vLLM, server mode, DAPO, Dr. GRPO, weighted rewards, judges, unsloth) are now documented as snippets in docs/grpo.md and no longer shipped as separate files. (#39, @cm8908)
  • Dependencies: TRL 1.17.1 and transformers 5.22.1 (obsolete workarounds removed), plus bumps to deepspeed, wandb, kernels, ruff and others. (#44)

Tests

Supply chain

  • This release has no SBOM attached: releases here are immutable, and the upload ran after publishing. The same code ships with an SBOM as v2.5.1.

Full Changelog: v2.4.0...v2.5.0

v2.4.0

Choose a tag to compare

@likejazz likejazz released this 03 Oct 07:23
Immutable release. Only release title and notes can be modified.

What's Changed

  • lint: fix remaining ruff findings and pin the rule set by @likejazz in #24
  • refactor: streamline file handling and improve code clarity by @likejazz in #25
  • fix: sync periodic checkpoint saves across ranks by @likejazz in #26
  • feat: add AsyncGRPO execution mode by @cm8908 in #28
  • feat: evaluate every checkpoint with Inspect and log the scores to W&B by @likejazz in #30
  • fix: parse repeated and valueless --config arguments 🤖🤖🤖 by @Momoyeyu in #32
  • feat: honor dataset weights in DPO and GRPO 🤖🤖🤖 by @Momoyeyu in #33

Full Changelog: v2.3.0...v2.4.0

v2.3.0

Choose a tag to compare

@likejazz likejazz released this 15 Sep 08:19
Immutable release. Only release title and notes can be modified.

What's Changed

  • Fix "ZoRO-3" typo in README Multi-GPUs section by @likejazz in #13
  • docs: add a Development section (ruff + pytest) to README by @likejazz in #15
  • fix: fail when --config points to a missing YAML by @HarshRajSinghania in #14
  • docs: consolidate README intro and describe the four trainers by @likejazz in #17
  • ci: add GitHub Actions workflow by @Yyash64 in #16
  • ci: fix format check failure and polish CI workflow by @likejazz in #18
  • utils: add dataset validator for prompt-only (GRPO/Distill) datasets by @likejazz in #20
  • ci: enable the lint gate and tidy the workflow by @likejazz in #21
  • ci: track master after default branch rename by @likejazz in #22

New Contributors

Full Changelog: v2.2.0...v2.3.0

v2.2.0

Choose a tag to compare

@likejazz likejazz released this 12 Sep 03:18
Immutable release. Only release title and notes can be modified.

What's Changed

  • feat: add On-Policy Distillation support (distill.py) by @Matlab7 in #1
  • Add dynamic sampling for zero-advantage GRPO groups by @cm8908 in #2
  • fix: stale logging and output-directory tests by @cm8908 in #3
  • fix(grpo): make dynamic sampling fire at pd > 1 and remove the wrong fallback by @cm8908 in #4

New Contributors

Full Changelog: https://github.com/dnotitia/dna-factory/commits/v2.2.0