## pytorch/torchtitan — ciflow/8gpu/4202…ciflow/8gpu/4203

_6 commits._

### Features
- **Add pluggable FSDP-managed MXFP8 linear training** (8ea9eff)
- **Revert "Add optimized MXFP8 training paths"** (1fe1294)
- **Add optimized MXFP8 training paths** (aebd04e)
- **Drop Python 3.10 support, require Python 3.11+ (#4180)** (233c22f)

### Backend
- **Make MXFP8 weight storage contract explicit** (662b195)
- **spmd_types as default backend (#4085)** (5ab3a0f)

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