onnxruntime
[CUDA] Support attention_bias in GroupQueryAttention via the unfused path
#29525

Merged

Commits

[CUDA] Support attention_bias in GroupQueryAttention via the unfused path

jiangzhuo committed 12 days ago
Address review: consistent symbolic dim name, wrap long line

jiangzhuo committed 12 days ago
Address review: bias rank guard, CUDA skip gate, broadcast-flag test coverage

jiangzhuo committed 11 days ago
Lint: RUF005 tuple unpacking, ruff-format line wrap

jiangzhuo committed 10 days ago
Add C++ attention_bias parity test (CUDA vs CPU)

jiangzhuo committed 7 days ago