Helper function to calculate strides for broadcasting

Dumps the JSON object to a formatted string.

Enable tracking of min/max values to update scale and zero_point.

Execute a single, stateless shell command on device. This command will NOT maintain context like environment variables. For s

FIXME: deprecated, context has no device

! \brief Gets a list of devices for this platform. * * Wraps clGetDeviceIDs(). */

Get canonical combining class for a codepoint using existing flags data

Get list of connected devices

export initQnnBackend function to initialize QNN backend

* Performs optimized matrix multiplication on CPU. * * This subroutine may compute C = Aᵀ * B with column major ordering. * Despite its name, this

* Register a new conversation template */

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

Rotates half the hidden dims of the input.

Rotates half the hidden dims of the input.

Rotates half the hidden dims of the input.

Safe sofmax for fp32. Not optimized for stride!=1 situation. When stride is set to 1, this function will utilize vexp1_fast_fp32 method to accelerate

adapted from arm limited optimized routine the maximum error is 1.45358 plus 0.5 ulps numbers above 88.38 will flush to infinity numbers beneath -103.

Helper to fill ATB tensor with custom shape (for reshape without copy)