fix typo

Signed-off-by: sohams <[email protected]>

tutorials/cuda_mps/README.md

@@ -182,9 +182,9 @@ We look at the performance benefits of MPS by varying the number of instances an
 
 We fix the number of parallel inferences to 7, number of runs to 10 and number of messages to 1000 and vary the number of instances from 3 to 7 using the `-i` parameter. Please refer to [Performance Benchmark Setup](#performance-benchmark-setup) for benchmarking commands.
 
-The graph below shows the maximum end-to-end latency of model benchmarking application with and without CUDA MPS, where the active thread percentage was set to `80/(number of instances)`. For example, for 5 instances, we set the active thread percentage to `80/5 = 16`. By provisioning resources this way, we leave some resouces idle in case a client should require to use it. Please refer to [CUDA MPS Resource Provisioning](https://docs.nvidia.com/deploy/mps/#volta-mps-execution-resource-provisioning) for more details regarding this.
+The graph below shows the maximum end-to-end latency of model benchmarking application with and without CUDA MPS, where the active thread percentage was set to `80/(number of instances)`. For example, for 5 instances, we set the active thread percentage to `80/5 = 16`. By provisioning resources this way, we leave some resources idle in case a client should require to use it. Please refer to [CUDA MPS Resource Provisioning](https://docs.nvidia.com/deploy/mps/#volta-mps-execution-resource-provisioning) for more details regarding this.
 
-The graph is missing a bar for the case of 7 instances and 7 parallel inferences as we were unable to get the baseline to execute. However, we were able to run when MPS was enabled, highlighting the advantage of using MPS for large workloads. We see that the maximum end-to-end latency improves when MPS is enabled and the improvement is more pronounced as the number of instances increases. This is beacuse, as the number of concurrent processes increases, MPS confines CUDA workloads to a certain predefined set of SMs. MPS combines multiple CUDA contexts from multiple processes into one, while simultaneously running them together.
+The graph is missing a bar for the case of 7 instances and 7 parallel inferences as we were unable to get the baseline to execute. However, we were able to run when MPS was enabled, highlighting the advantage of using MPS for large workloads. We see that the maximum end-to-end latency improves when MPS is enabled and the improvement is more pronounced as the number of instances increases. This is because, as the number of concurrent processes increases, MPS confines CUDA workloads to a certain predefined set of SMs. MPS combines multiple CUDA contexts from multiple processes into one, while simultaneously running them together.
 It reduces the number of context switches and related inferences, resulting in improved GPU utilization.