May d410a5652d fix(train): align DDP no_sync with gradient accumulation (#3472) ...

Synchronize DDP/FSDP gradients on the final microbatch of each accumulation window.

Verified with a two-rank CPU/Gloo reproduction, py_compile, diff checks, and a clean synthetic merge against current main.

2026-08-05 11:32:24 +08:00

..

__init__.py

funasr2

2023-12-15 23:46:41 +08:00

add_gradient_noise.py

funasr2

2023-12-15 23:46:41 +08:00

average_nbest_models.py

Dev kws (#2105)

2024-09-25 15:10:50 +08:00

device_funcs.py

Dev gzf exp (#1654)

2024-04-24 16:03:38 +08:00

forward_adaptor.py

docs: add docstrings to 1950 functions across 276 files for API documentation

2026-05-21 07:52:52 +00:00

initialize.py

Dev gzf exp (#1654)

2024-04-24 16:03:38 +08:00

load_pretrained_model.py

fix: disable timestamps when CTC weights are missing (#3211)

2026-07-13 23:57:17 +08:00

model_summary.py

docs: add docstrings to 1950 functions across 276 files for API documentation

2026-05-21 07:52:52 +00:00

recursive_op.py

docs: add docstrings to 1950 functions across 276 files for API documentation

2026-05-21 07:52:52 +00:00

set_all_random_seed.py

docs: add docstrings to 1950 functions across 276 files for API documentation

2026-05-21 07:52:52 +00:00

trainer_ds.py

fix(train): align DDP no_sync with gradient accumulation (#3472)

2026-08-05 11:32:24 +08:00

trainer.py

fix(trainer): avoid KeyError when a checkpoint is saved at an unvalidated step (#3455)

2026-08-03 14:19:25 +08:00