Changelog

v0.8.0 (Sep. 7, 2023)

Improve performance.
[Removed] Remove num_unit_steps parameter. This parameter is now automatically adjusted internally.

Improve performance greatly.
Support NVIDIA A100 GPU through a newly added service contract plan, which speeds up large-scale problems and accepts fully connected 131,072 bit-problems at the largest (twice as currently running V100 GPU).
Support multi GPU computation. Add num_gpus parameter to indicate the number of GPUs for annealing.
Add runtime version information to execution parameter outputs.

Improve performance.
Improve precision of internal calculations.
Add support of penalty function inputs.
- Automatically calibrate multipliers of penalty functions.
- Return feasibilities parameter which shows whether each solution is feasible.
Extend execution parameter outputs
- timeout to show the input timeout value.
- num_iterations to indicate the number of iterations of optimization until timeout.
- penalty_calibration to show whether automatic calibration of penalty functions is enabled.
- penalty_multipliers to show (calibrated) multipliers of penalty functions.
Extend polynomial parameter to accept a constant value.

Improves performance for problems with variables that do not contain interactions.

Improve performance.
Accept upper and lower triangular matrix for matrix key of the request parameter.
Accept encoded request and response body compressed by deflate or gzip. See example commands.
Set the value of num_unit_steps to 10 as default.

Improve performance.
Increase maximum problem size.
- Fully connected (matrix): 65,536 bits
- Sparse (polynomial): 4,294,967,296 elements