Exponential function. This function is exact to rounding for all rounding modes. See also:** Documentation for [std::exp](https://en.cppreference.com

Power function. This function may be 1 ULP off the correctly rounded exact result for any rounding mode in ~0.00025% of inputs. See also:** Documenta

! @brief get next character from the input This function provides the interface to the used input adapter. It does not throw in case the

Fixed point binary logarithm. This uses the BKM algorithm in L-mode. \param m mantissa in [1,2) as Q1.30 \param n number of iterations (at most 32) \r

Verify that the required weights are set. Raises: ValueError: If a required weight is not set in the specification.

@brief return whether value is a binary array @sa https://json.nlohmann.me/api/basic_json/is_binary/

Registers a source vocabulary of tokens. Arguments: tokens: List of source tokens.

Registers a target vocabulary of tokens. Arguments: tokens: List of target tokens.

Platform-independent sign mask. \param arg integer value in two's complement \retval -1 if \a arg negative \retval 0 if \a arg positive

Helper function to call :meth:`ctranslate2.converters.Converter.convert` with the parsed command line options. Arguments: a

Adds common conversion options to the command line parser. Arguments: parser: Command line argument parser.

64-bit division. \param x upper 32 bit of dividend \param y divisor \param s variable to store sticky bit for rounding \return (\a x << 32) / \a y

Multiply by power of two. This function is exact to rounding for all rounding modes. See also:** Documentation for [std::ldexp](https://en.cppreferen

64-bit multiplication. \param x first factor \param y second factor \return upper 32 bit of \a x * \a y rounded to nearest

! @brief internal implementation of the serialization function This function is called by the public member function dump and organizes t

Fixed point binary exponential. This uses the BKM algorithm in E-mode. \param m exponent in [0,1) as Q0.31 \param n number of iterations (at most 32)

Recursively applies some optimizations to this layer: * Alias variables with the same shape and value. * Quantize weights. A

activation_type(x + bias + residual)

Check and signal for any NaN. \param x first half-precision value to check \param y second half-precision value to check \retval true if either \a x o

can't use conversion operator because of ambiguity