Model components / Fine-tuning code

torchtune

torchtune (pytorch/torchtune), BSD-3-Clause, confirmed live June 2026 but DEPRECATED; repo states 'torchtune is no longer actively maintained; development wound down in 2025.' Provided SFT, knowledge distillation, DPO/PPO/GRPO, QAT; single/multi-device/multi-node; FSDP2; HF Hub + W&B integration.

Openness

5 high confidence

5.0

license: BSD-3-Clause(OSI)
source: public(pytorch/torchtune)
governance: PyTorch Foundation/Meta
core-gated: ungated

Fully OSI-licensed (BSD-3-Clause), full source public; openness unaffected by the project being wound down.

https://github.com/pytorch/torchtune recorded 2026-06-04
BSD-3-Clause license; PyTorch post-training library

Adoption

3 medium confidence

3.0

~370K PyPI downloads last month, but the project is officially deprecated (development wound down in 2025), so usage is on a declining trajectory; ~5.8K stars. Level 3 on current download volume with the stale caveat flagged.

https://pypistats.org/packages/torchtune recorded 2026-06-04
Downloads last month: 370,048
https://github.com/pytorch/torchtune recorded 2026-06-04
'no longer actively maintained; development wound down in 2025'; ~5.8K stars

Capability

3 medium confidence

3.0

Competent native-PyTorch method/parallelism coverage but development halted in 2025, so it no longer tracks the moving frontier, C3.

https://github.com/pytorch/torchtune recorded 2026-06-04
SFT/distillation/DPO/PPO/GRPO/QAT, FSDP2, single/multi-device/multi-node

Unchanged since 2026-07-30 (last edited, not re-checked)