@spec print_interpreter_state(reference()) :: nil | {:error, String.t()}

Prints a dump of what tensors and what nodes are in the interpreter.

Note that this function directly prints to stdout. It answers nil once it has, or {:error, reason} when another process is using the interpreter or it belongs to one, which used to be swallowed and reported as nil.

@spec reset_variable_tensor(
  %TFLiteElixir.TFLiteTensor{
    index: term(),
    name: term(),
    quantization_params: term(),
    reference: term(),
    shape: term(),
    shape_signature: term(),
    sparsity_params: term(),
    type: term()
  }
  | reference()
) :: :ok | {:error, String.t()}

Resets a variable tensor to the default value.

@spec source_tree() :: :litert

Which source tree the loaded shared object was built from. Answers :litert.

There is no other answer: the C++ behind it names a type only LiteRT's schema defines, so a binary built from anything else does not compile, and a release from before the move has no such function at all. Worth asking in a test rather than trusting the build, because a stale precompiled artifact looks exactly like a fresh one from the outside.

@spec tensorflow_version() :: String.t()

The TensorFlow release this build pulled in, for example "2.21.0-rc0".

TensorFlow is not where the runtime comes from any more: LiteRT reaches into it for compiler/mlir/lite, TSL and XLA. Worth having when something reads wrong, not for matching a plugin against.

@spec tflite_extension_apis_version() :: String.t()

The version of the extension APIs: c_api_opaque.h, common.h, builtin_op_data.h and builtin_ops.h.

Narrower in scope than tflite_runtime_version/0 but derived from the same stale number, so the same caveat applies.

@spec tflite_runtime_version() :: String.t()

The runtime version string lite/version.h carries.

Hand-maintained upstream and forgotten: the 2.21.0 tree still says "2.19.0". It only applies when the build system injects nothing, which Bazel does and CMake does not, so two builds from different releases are indistinguishable through this value. tflite_version/0 is the one to match a plugin against.

@spec tflite_schema_version() :: integer()

The major schema version this runtime reads model files at.

Unlike the version strings this one is real: it is defined next to the schema it describes. A model serialised at a different schema version may not load.

@spec tflite_version() :: String.t()

The version of the TfLite sources this was built from, for example "2.2.0".

Since the runtime moved to LiteRT's tflite subtree this is LiteRT's version, not TensorFlow's. The two are separate version lines and the numbers are not comparable: LiteRT's 2.2.0 is newer than TensorFlow's 2.21.0, not older. For the TensorFlow release the build borrowed from, see tensorflow_version/0.

This is the number a delegate plugin has to match. A plugin loaded through TFLiteElixir.Delegate.external/1 must have been built from the same release; upstream offers no binary stable delegate interface, so a mismatch is undefined behaviour rather than an error.

@spec xnnpack_max_tensor_dims() :: integer() | nil

How many dimensions XNNPACK will delegate a tensor with, or nil where this build carries no XNNPACK.

A tensor already wider than this was refused by the delegate to begin with, was therefore never delegated, and can still be reshaped freely. The armv6 and armv7l builds answer nil, where nothing is refused.