Mojo function
concat_q_nope_proj_rope
concat_q_nope_proj_rope[dtype: DType, target: StringSlice[StaticConstantOrigin] = "cpu"](output: TileTensor[dtype, origin, linear_idx_type=linear_idx_type, element_shape_types=element_shape_types], q_nope_proj: TileTensor[dtype, origin, linear_idx_type=linear_idx_type, element_shape_types=element_shape_types], q_rope: TileTensor[dtype, origin, linear_idx_type=linear_idx_type, element_shape_types=element_shape_types], ctx: DeviceContext)
Concatenate projected q_nope and q_rope into a single q tensor.
Expects q_nope_proj shape [H, S, D] and q_rope shape [S, H, D_rope], and writes output shape [S, H, D_proj + D_rope].
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!