ViperEkura bcaa2d1ae0 fix: FSDP unwrap_model collective op and None guard ...

- unshard() and full_tensor() are collective ops, all ranks must participate
- Old code returned None on non-rank-0 before calling unshard, causing deadlock
- Fix: all ranks unshard/full_tensor, only rank-0 keeps the result
- Move create_ref_model to parallel/utils.py, accept executor+model directly
- Guard create_ref_model and sync_old_model against None on non-rank-0

2026-07-29 23:41:10 +08:00

..

config

feat: add torch.compile CLI option for training

2026-07-29 22:06:51 +08:00

dataset

refactor: remove H5 storage backend in favor of mmap bin

2026-07-29 12:50:27 +08:00

extension

feat: add attention dimension dispatch

2026-07-21 12:52:45 +08:00

inference

perf: use Rust-native DecodeStream for O(n) streaming decode

2026-07-28 14:32:10 +08:00

model

refactor : align config field names with Hugging Face

2026-07-20 22:05:31 +08:00

parallel

fix: FSDP unwrap_model collective op and None guard

2026-07-29 23:41:10 +08:00

preprocessing

refactor: remove H5 storage backend in favor of mmap bin

2026-07-29 12:50:27 +08:00

serialization

refactor: remove H5 storage backend in favor of mmap bin

2026-07-29 12:50:27 +08:00

tokenize

fix: make ChatTemplate picklable for spawn multiprocessing

2026-07-29 13:24:13 +08:00

trainer

fix: FSDP unwrap_model collective op and None guard

2026-07-29 23:41:10 +08:00

__init__.py

refactor: remove H5 storage backend in favor of mmap bin

2026-07-29 12:50:27 +08:00

factory.py

fix: make FSDP2 executor work with ABC+Generic model hierarchy

2026-07-23 16:11:02 +08:00

protocols.py

refactor: 重构训练后端为 Executor 模式

2026-05-24 20:35:44 +08:00

signal_handler.py

refactor: move signal_handler from parallel/ to top-level for broader reuse

2026-07-28 10:36:17 +08:00