onnxruntime
Fix LinearAttention output shape inference for standard GQA (q > kv)
#29892
Merged
Merged