ViperEkura 75411ce0cc fix: skip CUDA rotary kernel when grad is enabled ...

- apply_rotary_emb now checks torch.is_grad_enabled() before dispatching to CUDA kernel
- Training (grad enabled) uses torch complex multiply path which supports autograd backward
- Inference (inference_mode/no_grad) uses CUDA kernel as before
- Without this fix, training backward would crash — the CUDA kernel has no autograd backward()

2026-07-31 15:43:15 +08:00

..

config

feat: add field and model validators to config classes

2026-07-30 08:41:14 +08:00

dataset

refactor: remove H5 storage backend in favor of mmap bin

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

extension

fix: skip CUDA rotary kernel when grad is enabled

2026-07-31 15:43:15 +08:00

inference

perf: reduce decode overhead in scheduler and executor

2026-07-31 14:50:16 +08:00

model

perf: add fused CUDA rotary embedding kernel

2026-07-31 15:27:31 +08:00

parallel

fix: broadcast ref/old model state_dict for FSDP

2026-07-31 08:32:22 +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

refactor: simplify BaseFactory and separate ModelFactory from AutoModel

2026-07-30 09:38:20 +08:00

trainer

fix: broadcast ref/old model state_dict for FSDP

2026-07-31 08:32:22 +08:00

__init__.py

refactor: remove H5 storage backend in favor of mmap bin

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

factory.py

refactor: simplify BaseFactory and separate ModelFactory from AutoModel

2026-07-30 09:38:20 +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