Backend: Some Tensors Not Going Through set_tensor

[taronaeo]

I noticed that some of my tensors such as blk.0.attn_q.weight are not going through the set_tensor function while tensors with the naming scheme of the backend such as zDNN#attn_norm-0#0 are going through correctly. This is causing quite a problem for my backend during inference on operations such as GGML_OP_MUL_MAT because the abovementioned 2 tensors would come in and only 1 of them are initialised correctly via set_tensor.

As a sanity check, I have ensured that my backend is only broadcasting buffer types it can support i.e.,

static bool ggml_backend_zdnn_device_supports_buft(ggml_backend_dev_t dev, ggml_backend_buffer_type_t buft) {
    return
        buft->iface.get_name == ggml_backend_zdnn_buffer_type_get_name ||  // "zDNN"
        buft->iface.get_name == ggml_backend_zdnn_buffer_from_ptr_type_get_name;  // "zDNN_Mapped"

    GGML_UNUSED(dev);
}

Pardon my ignorance if any. I have been trying to debug this for hours and still have no idea what is wrong here.

Any help would be greatly appreciated!