Remove llava from ci_expected_accuracy as it's flaky (#121322)
https://github.com/pytorch/pytorch/pull/121029 added it into the CI but the test is flaky on hud. It alternates between fail_accuracy and fail_to_run
Pull Request resolved: https://github.com/pytorch/pytorch/pull/121322
Approved by: https://github.com/desertfire
diff --git a/benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_torchbench_inference.csv b/benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_torchbench_inference.csv
index dbfbcda..cca1164 100644
--- a/benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_torchbench_inference.csv
+++ b/benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_torchbench_inference.csv
@@ -178,10 +178,6 @@
-llava,fail_accuracy,0
-
-
-
maml,pass_due_to_skip,0
diff --git a/benchmarks/dynamo/ci_expected_accuracy/dynamic_cpu_inductor_torchbench_inference.csv b/benchmarks/dynamo/ci_expected_accuracy/dynamic_cpu_inductor_torchbench_inference.csv
index ab77e69..4753fb4 100644
--- a/benchmarks/dynamo/ci_expected_accuracy/dynamic_cpu_inductor_torchbench_inference.csv
+++ b/benchmarks/dynamo/ci_expected_accuracy/dynamic_cpu_inductor_torchbench_inference.csv
@@ -138,10 +138,6 @@
-llava,fail_accuracy,0
-
-
-
maml,pass_due_to_skip,0
diff --git a/benchmarks/dynamo/torchbench.yaml b/benchmarks/dynamo/torchbench.yaml
index fda7af5..240881b 100644
--- a/benchmarks/dynamo/torchbench.yaml
+++ b/benchmarks/dynamo/torchbench.yaml
@@ -182,6 +182,7 @@
# works on cuda, accuracy failure on cpu
- hf_Whisper
- stable_diffusion_text_encoder
+ - llava
cuda: []
