Quick Start#

This chapter gets you up and running with Kaiwu-PyTorch-Plugin through short, self-contained examples.

1. Basic Examples#

1.1 Restricted Boltzmann Machine (RBM)#

The snippet below shows basic training with the RestrictedBoltzmannMachine class. You can set the number of visible and hidden units and optionally supply custom initial values for the quadratic and linear terms.

import torch

from torch.optim import SGD
import kaiwu as kw
from kaiwu.torch_plugin import RestrictedBoltzmannMachine
from kaiwu.classical import SimulatedAnnealingOptimizer
from kaiwu.cim import CIMOptimizer, PrecisionReducer

from kaiwu.cim import CIMOptimizer, PrecisionReducer

# 添加licence认证
# print("User ID:", os.getenv("USER_ID"), "SDK Code:", os.getenv("SDK_CODE"))
# kw.license.init(os.getenv("USER_ID"), os.getenv("SDK_CODE"))


if __name__ == "__main__":
    NUM_READS = 1
    SAMPLE_SIZE = 1
    USE_CIM = False

    if USE_CIM:
        kw.common.CheckpointManager.save_dir = "./tmp"
        sampler = CIMOptimizer(task_name="test_kpp", wait=True)
        sampler = PrecisionReducer(
            sampler,
            precision=8,
            truncated_precision=10,
            target_bits=550,
            only_feasible_solution=False,
        )
    else:
        sampler = SimulatedAnnealingOptimizer()
    num_nodes = 5
    num_visible = 2
    x = 1.0 * torch.randint(0, 2, (SAMPLE_SIZE, num_visible))

    # Instantiate the model
    rbm = RestrictedBoltzmannMachine(
        num_visible,
        num_nodes - num_visible,
        quadratic_coef=torch.FloatTensor(
            [
                [2, -3, 0],
                [-1, 2, 0],
            ]
        ),
        linear_bias=torch.FloatTensor([1, 1, 0, -1, 2]),
    )
    # Instantiate the optimizer
    opt_rbm = SGD(rbm.parameters())

    # Example of one iteration in a training loop
    # Generate a sample set from the model
    x = rbm.get_hidden(x, bernoulli=True)
    s = rbm.sample(sampler)
    opt_rbm.zero_grad()
    # Compute the objective---this objective yields the same gradient as the negative
    # log likelihood of the model
    objective = rbm.objective(x, s)
    # Update model weights with a step of stochastic gradient descent
    objective.backward()

1.2 Boltzmann Machine (BM)#

The next example demonstrates the BoltzmannMachine class:

import torch

from torch.optim import SGD
from kaiwu.classical import SimulatedAnnealingOptimizer
from kaiwu.torch_plugin import BoltzmannMachine

#  这里添加licence认证


if __name__ == "__main__":
    SAMPLE_SIZE = 5

    sampler = SimulatedAnnealingOptimizer(alpha=0.99, size_limit=5)
    sample_kwargs = {}
    num_nodes = 5
    num_visible = 2
    x = 1.0 * torch.randint(0, 2, (SAMPLE_SIZE, num_visible))

    # Instantiate the model
    rbm = BoltzmannMachine(num_nodes)

    # Instantiate the optimizer
    opt_rbm = SGD(rbm.parameters())

    # Example of one iteration in a training loop
    # Generate a sample set from the model

    x = rbm.condition_sample(sampler, x)
    s = rbm.sample(sampler)
    opt_rbm.zero_grad()
    # Compute the objective---this objective yields the same gradient as the negative
    # log likelihood of the model
    objective = rbm.objective(x, s)
    # Backpropgate gradients
    print("call backward")
    objective.backward()
    print("after backward")
    # Update model weights with a step of stochastic gradient descent
    opt_rbm.step()
    print(objective)

3. Using Different Samplers#

Kaiwu SDK offers several samplers; pick the one that best fits your needs:

 from kaiwu.classical import SimulatedAnnealingOptimizer

 # Simulated-annealing optimizer (recommended for most scenarios)
 sampler_sa = SimulatedAnnealingOptimizer()

 # To use the quantum sampler (requires real-machine access)
 # from kaiwu.cim import CIMOptimizer

4. Next Steps#

Congratulations—you have finished the quick-start! Where to go next:

On this page