How to access the ggml_tensor data?

[ebeyabraham]

Hi, this may be a silly question but how can I access the data stored in a ggml_tensor? For example, in examples/gpt-2/main.cpp when I try printing some vector as below I get a segmentation fault:

  for (int i = 0; i < 5; ++i) {
      std::cout << ((float*) inpL->data)[i];
  }

The above method used to work for me in a previous version of ggml, but do longer works on the current master.

[slaren]

The gpt-2 example was recently updated to use ggml-alloc, and one of the consequences is that tensors don't have data allocated (ie. tensor->data is NULL) until explicitly allocated with ggml_allocr_alloc or ggml_allocr_alloc_graph.

You shouldn't need to access the tensor data unless you are trying to set the inputs (in which case you can use ggml_allocr_alloc to allocate the tensor immediately), or you are trying to retrieve outputs, in which case all the tensors in the graph would have already been allocated by ggml_allocr_alloc_graph. What are you trying to do exactly?