Check if this is the 'max context' doc span for the token.

Performs invalid character removal and whitespace cleanup on text.

Compute softmax probability over raw logits.

Checks whether CP is the codepoint of a CJK character.

Checks whether `chars` is a control character.

Checks whether `chars` is a punctuation character.

Checks whether `chars` is a whitespace character.

Splits punctuation on a piece of text.

Strips accents from a piece of text.

Adds whitespace around any CJK character.

Add the attention layer

Create the bert model

Build a dependency graph of the headers.

Return the list of tokens from the doc text

Extract function names and versions from CUDA PFN type definitions.

Find all C/C++ header files in the given directory and subdirectories.

Recursively find and process function declarations.

BERT demo needs a separate engine building path to generate calibration cache. This is because we need to configure SLN and MHA plugins in FP

Generate typedef string for complex types.

Generate typedef name for complex types.

Project the tokenized prediction back to the original text.

Extract symbols from the library file using nm or dumpbin.

Convert a shell-style glob pattern to a regex pattern.

Check if the type is a function pointer or array.

Check if an include is part of our project's headers.

Load the weights from the onnx checkpoint

Load the weights from the pytorch checkpoint

Load the weights from the tensorflow checkpoint

Merge headers, preserving formatting, comments, and external includes.

Parse a function declaration and generate corresponding code.

Parse the source file and generate code.

Preprocess the file using clang with the specified include paths. Args: file_path: Path to the file to preprocess include_pat

Create the squad output

Tokenizes a piece of text.

Sort headers in topological order to resolve dependencies.

Add the transformer layer

Write the generated code to output files.