刘鑫 b1584aec7c fix: stabilize CPU SDPA mask broadcasting ...

Use an explicit broadcastable attention mask shape during MiniCPM incremental decoding so CPU runtimes avoid a PyTorch SDPA dimension error without changing attention semantics.

Made-with: Cursor

2026-04-13 15:38:53 +08:00

..

model

fix: close file handles in from_local() config loading

2026-04-11 00:01:14 -04:00

modules

fix: stabilize CPU SDPA mask broadcasting

2026-04-13 15:38:53 +08:00

training

update finetuning pipeline and runtime device handling

2026-04-11 11:08:50 +08:00

utils

update voxcpm2

2026-03-31 11:50:37 +08:00

__init__.py

init

2025-09-16 11:46:47 +08:00

cli.py

update finetuning pipeline and runtime device handling

2026-04-11 11:08:50 +08:00

core.py

update finetuning pipeline and runtime device handling

2026-04-11 11:08:50 +08:00

zipenhancer.py

update voxcpm2

2026-03-31 11:50:37 +08:00