Replace ones with rand to workaround the crash from sdpa kernel

Author: Github Executorch

.github/workflows/android-perf.yml

@@ -86,7 +86,7 @@ jobs:
           # Separate default values from the workflow dispatch. To ensure defaults are accessible
           # during scheduled runs and to provide flexibility for different defaults between
           # on-demand and periodic benchmarking.
-          CRON_DEFAULT_MODELS: ${{ github.event_name == 'schedule' && 'stories110M,dl3,mv3,mv2,ic4,ic3,vit,meta-llama/Llama-3.2-1B' || 'stories110M' }}
+          CRON_DEFAULT_MODELS: ${{ github.event_name == 'schedule' && 'stories110M,dl3,mv3,mv2,ic4,ic3,vit' || 'stories110M' }}
           CRON_DEFAULT_DEVICES: samsung_galaxy_s22
           CRON_DEFAULT_DELEGATES: ${{ github.event_name == 'schedule' && 'xnnpack,qnn' || 'xnnpack' }}
         run: |

.github/workflows/apple-perf.yml

@@ -76,7 +76,7 @@ jobs:
           # Separate default values from the workflow dispatch. To ensure defaults are accessible
           # during scheduled runs and to provide flexibility for different defaults between
           # on-demand and periodic benchmarking.
-          CRON_DEFAULT_MODELS: ${{ github.event_name == 'schedule' && 'stories110M,mv3,mv2,ic4,ic3,resnet50,edsr,mobilebert,w2l,meta-llama/Llama-3.2-1B' || 'stories110M' }}
+          CRON_DEFAULT_MODELS: ${{ github.event_name == 'schedule' && 'stories110M,mv3,mv2,ic4,ic3,resnet50,edsr,mobilebert,w2l' || 'stories110M' }}
           CRON_DEFAULT_DEVICES: apple_iphone_15
           CRON_DEFAULT_DELEGATES: ${{ github.event_name == 'schedule' && 'xnnpack,coreml,mps' || 'xnnpack' }}
         run: |

extension/benchmark/apple/Benchmark/Tests/GenericTests.mm

@@ -78,7 +78,7 @@ @implementation GenericTests
 
           const auto sizes = tensor_meta->sizes();
           tensors.emplace_back(
-              ones({sizes.begin(), sizes.end()}, tensor_meta->scalar_type()));
+              rand({sizes.begin(), sizes.end()}, tensor_meta->scalar_type()));
           XCTAssertEqual(module->set_input(tensors.back(), index), Error::Ok);
         } break;
         default:

extension/benchmark/apple/Benchmark/Tests/LLaMA/LLaMATests.mm

@@ -87,7 +87,7 @@ @implementation LLaMATests
                               tokensPerSecondMetric.tokenCount = 0;
                               const auto status = runner->generate(
                                   "Once upon a time",
-                                  128,
+                                  50,
                                   [=](const std::string &token) {
                                     tokensPerSecondMetric.tokenCount++;
                                   },