Michael Bartling

Models interface should push and pop default allocator context on eval

3

To support multiple allocator contexts in the tensor destructor, model interface should push and pop current allocator pointers to a queue in the default current context else risks failed destruction

Add a couple more reference operators.

[Feature Request] Add shape checking to Placeholder Tensors

3

Right now there is no shape checking on Placeholder tensors, this means that invalid shapes will still propagate through the model and lead to weird errors.

dirent.h@10,2: #error "<dirent.h> not supported" on some platforms

2

need a smart way of getting around this.

Earlier we found that just removing std::string and using c string, we could shrink code size dramatically. Ultimately we need to roll out our own string class, here are a...

Update Tensor R/W interface

3

Proposed update for the Tensor read write interface: ```cpp // Currently template const T* read(size_t offset, size_t ele) { return (const T*)read(offset, ele); } ... const float input_min = in_min->read(0,...

Hey guys, this should be pretty low hanging fruit. https://www.tensorflow.org/api_docs/python/tf/negative OPNAME: "Neg"

init_env should be moved to code gen

1

Right now `init_env()` is platform specific. This needs to be either made into a class with empty default, or moved completely into code_gen

It would be nice to have ``` explicit Tensor( size_type count ); ``` With similar behavior to ``` explicit std::vector( size_type count ); ``` Basically, it allows you to construct...