]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
CANN: Resolve soft_max precision issue (llama/15730)
authorhipudding <redacted>
Tue, 2 Sep 2025 09:12:37 +0000 (17:12 +0800)
committerGeorgi Gerganov <redacted>
Sat, 20 Sep 2025 10:42:47 +0000 (13:42 +0300)
commit5aee53c40fa42c238886dc003b9d78436bf250e6
tree64e4d413209f96d74704f50c97b98aa60df5298e
parent1e03aa66f795a9d3890c0165214772da3a710f3d
CANN: Resolve soft_max precision issue (llama/15730)

Previously, the slope tensor was set to fp16 to improve efficiency.
While this worked correctly in FA, it caused precision issues in soft_max.
This change applies different data types for different operators
to balance both accuracy and performance.
ggml/src/ggml-cann/aclnn_ops.cpp