AWS braket documentation change
Summary
Minor grammatical and clarity improvements: simplified phrasing in gate decomposition instructions, replaced 'leverage' with 'use' in data parallelism section, and clarified 'i.e.' as 'that is' in a cost example.
Security assessment
Changes are purely editorial, focusing on readability and phrasing rather than addressing security vulnerabilities or documenting security features. No security-related terminology, configurations, or warnings were added or modified.
Diff
diff --git a/braket/latest/developerguide/pennylane-embedded-simulators.md b/braket/latest/developerguide/pennylane-embedded-simulators.md index 1507113a6..07358b569 100644 --- a//braket/latest/developerguide/pennylane-embedded-simulators.md +++ b//braket/latest/developerguide/pennylane-embedded-simulators.md @@ -20 +20 @@ Consider the Quantum Approximate Optimization Algorithm (QAOA) examples from thi -In this page, let’s compare the two embedded PennyLane state vector simulators `lightning.qubit` (which is CPU-based) and `lightning.gpu` (which is GPU-based). You’ll need to provide the simulators with some custom gate decompositions in order to compute various gradients. +In this page, let’s compare the two embedded PennyLane state vector simulators `lightning.qubit` (which is CPU-based) and `lightning.gpu` (which is GPU-based). You’ll need to provide the simulators with some custom gate decompositions to compute various gradients. @@ -106 +106 @@ If your workload type is quantum machine learning (QML) that trains on datasets, -Because the loss from one data point does not depend on other data points, the losses can be evaluated in parallel! Losses and gradients associated with different data points can be evaluated at the same time. This is known as data parallelism. With SageMaker’s distributed data parallel library, Amazon Braket Hybrid Jobs make it easier for you to leverage data parallelism to accelerate your training. +Because the loss from one data point does not depend on other data points, the losses can be evaluated in parallel! Losses and gradients associated with different data points can be evaluated at the same time. This is known as data parallelism. With SageMaker’s distributed data parallel library, Amazon Braket Hybrid Jobs make it easier for you to use data parallelism to accelerate your training. @@ -205 +205 @@ In the above hybrid job creation, `train_dp.py` is the modified algorithm script -Let’s compare the run time and cost in an example where when train a model with a 26-qubit quantum circuit for the binary classification problem mentioned above. The `ml.p3.16xlarge` instance used in this example costs $0.4692 per minute. Without data parallelism, it takes the simulator about 45 minutes to train the model for 1 epoch (i.e., over 208 data points) and it costs about $20. With data parallelism across 1 instance and 4 instances, it only takes 6 minutes and 1.5 minutes respectively, which translates to roughly $2.8 for both. By using data parallelism across 4 instances, you not only improve the run time by 30x, but also reduce costs by an order of magnitude! +Let’s compare the run time and cost in an example where when train a model with a 26-qubit quantum circuit for the binary classification problem mentioned above. The `ml.p3.16xlarge` instance used in this example costs $0.4692 per minute. Without data parallelism, it takes the simulator about 45 minutes to train the model for 1 epoch (that is, over 208 data points) and it costs about $20. With data parallelism across 1 instance and 4 instances, it only takes 6 minutes and 1.5 minutes respectively, which translates to roughly $2.8 for both. By using data parallelism across 4 instances, you not only improve the run time by 30x, but also reduce costs by an order of magnitude!