## deepspeedai/DeepSpeed — v0.19.1…v0.19.2

_27 commits._

### Features
- **Add Biren SUPA accelerator support (#8054)** (7ad4108)
- **Add AutoEP (#7938)** (7c3fbdc)
- **Add test coverage for Muon muon_lr/adam_lr overrides (#8047)** (3dc98de)
- **feat(zero): enable torch.func transforms on engine for ZeRO 0/1/2 (#8026)** (4370aa5)
- **Add engine.coalesce_grad_reduction() for ZeRO 1/2/3 multi-backward (#7992)** (60b242a)

### Fixes
- **Fix: ZenFlow Adam integration for updated PyTorch backward flow (#7759) (#7771)** (b214589)
- **fix(transformer): use correct stride in Transpose_Kernel shared memory indexing to eliminate bank conflicts (#8055)** (65e5f6f)
- **Fix DeepCompile ZeRO-1 grad target lifetime (#8036)** (11eeb7c)
- **Fix ZenFlow ZeRO-3 selective optimizer crash with parameter offload on nvme (#8042)** (28a196f)
- **Fix DeepCompile ZeRO-3 release parameter lifetime (#8032)** (3e486fe)
- **Version fix to unblock pypi (#8039)** (d5d118c)
- **Fix DeepCompile all-gather scheduler candidate selection (#8033)** (819af0e)
- **Fix full CI test isolation for ZeRO chmod and NVMe quantization tests (#8008)** (4570c50)
- **Fix PR-target workflow concurrency groups (#8017)** (b7aef4d)
- **fix(fp16): filter requires_grad in FP16 optimizer flat buffer init (#8029)** (01e3a28)

### Backend
- **Mixed-precision: per-policy param/buffer dtype cast (preserve fp32 buffers) (#8066)** (b919284)
- **Pass expected grad dtype to register_z3_param in ZeRO-3 release test (#8063)** (3a917de)
- **Consistent fp32 grads flow (#8056)** (56e7e0c)
- **zero3: invalidate coordinator trace on hook re-registration (#8043)** (87f31d5)
- **Remove AutoSP assertion against Transformers version (#8044)** (a14e4e1)
- **Normalize ZeRO-3 DeepCompile grad dtype before reduction (#8038)** (4f5fb83)
- **Avoid HF Hub access in CPU unit test setup (#8053)** (92cfacb)
- **Bump version after 0.19.1 release (#8040)** (334230f)
- **Simplify module_inject.transpose (#8028)** (1afa3a5)
- **Bump version (#8030)** (a0b98d3)
- **Keep required CI checks visible for ignored paths (#8019)** (77a249f)
- **Run AutoSP compile tests sequentially (#8020)** (2c8a007)

_Recap by [Repo Wrapped](https://repowrapped.com/gh/deepspeedai/DeepSpeed?utm_source=github-action)._