onnxruntime
6372a026
- Pass kv_num_heads explicitly to ApplyAttention
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
16 days ago
Pass kv_num_heads explicitly to ApplyAttention Drop the default arguments so every caller supplies kv_num_heads from the parameters produced by CheckInputs, instead of letting ApplyAttention fall back to num_heads when 0 is passed.
References
#31680 - Add support for grouped-query MultiHeadAttention for CPU and CUDA
Author
tianleiwu
Committer
tianleiwu
Parents
87101f58
Loading