Reenable llama benchmark (#100877)

Signed-off-by: Edward Z. Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/100877
Approved by: https://github.com/albanD

benchmarks/dynamo/ci_expected_accuracy/inductor_torchbench_dynamic_inference.csv

@@ -6,7 +6,6 @@ Super_SloMo,pass,0
 alexnet,pass,0
 attention_is_all_you_need_pytorch,pass,0
 dcgan,pass,0
-densenet121,pass,0
 dlrm,pass,0
 doctr_det_predictor,pass,3
 doctr_reco_predictor,pass,3
@@ -23,6 +22,7 @@ hf_GPT2,pass,0
 hf_Reformer,pass,26
 hf_T5_large,pass_due_to_skip,0
 lennard_jones,pass,0
+llama,pass,0
 maml_omniglot,pass,0
 mnasnet1_0,pass,0
 mobilenet_v2,pass,0
@@ -31,7 +31,6 @@ nvidia_deeprecommender,pass,0
 opacus_cifar10,pass,44
 phlippe_densenet,pass,0
 phlippe_resnet,pass,0
-pyhpc_isoneutral_mixing,fail_to_run,0
 pytorch_CycleGAN_and_pix2pix,pass,0
 pytorch_stargan,pass,0
 pytorch_unet,pass,0
@@ -41,7 +40,6 @@ resnet50,pass,0
 resnext50_32x4d,pass,0
 shufflenet_v2_x1_0,pass,0
 soft_actor_critic,pass,0
-speech_transformer,fail_to_run,0
 squeezenet1_1,pass,0
 timm_efficientnet,pass,0
 timm_regnet,pass,0

benchmarks/dynamo/ci_expected_accuracy/inductor_torchbench_dynamic_training.csv

@@ -5,7 +5,6 @@ Super_SloMo,pass,9
 alexnet,pass,9
 attention_is_all_you_need_pytorch,pass,9
 dcgan,pass,9
-densenet121,pass,9
 drq,pass,8
 fastNLP_Bert,pass,14
 functorch_dp_cifar10,pass,9
@@ -19,6 +18,7 @@ hf_GPT2,pass,8
 hf_Reformer,pass,45
 hf_T5_large,pass_due_to_skip,0
 lennard_jones,pass,9
+llama,eager_1st_run_fail,0
 maml_omniglot,pass,9
 mnasnet1_0,pass,9
 mobilenet_v2,pass,9
@@ -34,7 +34,6 @@ resnet50,pass,9
 resnext50_32x4d,pass,9
 shufflenet_v2_x1_0,pass,9
 soft_actor_critic,pass,8
-speech_transformer,fail_to_run,4
 squeezenet1_1,pass,9
 timm_efficientnet,pass,9
 timm_regnet,pass,9

benchmarks/dynamo/ci_expected_accuracy/inductor_torchbench_inference.csv

@@ -6,7 +6,6 @@ Super_SloMo,pass,0
 alexnet,pass,0
 attention_is_all_you_need_pytorch,pass,0
 dcgan,pass,0
-densenet121,pass,0
 dlrm,pass,0
 doctr_det_predictor,pass,3
 doctr_reco_predictor,pass,3
@@ -23,6 +22,7 @@ hf_GPT2,pass,0
 hf_Reformer,pass,26
 hf_T5_large,pass_due_to_skip,0
 lennard_jones,pass,0
+llama,pass,0
 maml_omniglot,pass,0
 mnasnet1_0,pass,0
 mobilenet_v2,pass,0

benchmarks/dynamo/ci_expected_accuracy/inductor_torchbench_training.csv

@@ -5,7 +5,6 @@ Super_SloMo,pass,9
 alexnet,pass,9
 attention_is_all_you_need_pytorch,pass,9
 dcgan,pass,9
-densenet121,pass,9
 drq,pass,8
 fastNLP_Bert,pass,14
 functorch_dp_cifar10,pass,9
@@ -19,13 +18,13 @@ hf_GPT2,pass,8
 hf_Reformer,pass,67
 hf_T5_large,pass_due_to_skip,0
 lennard_jones,pass,9
+llama,eager_1st_run_fail,0
 maml_omniglot,pass,9
 mnasnet1_0,pass,9
 mobilenet_v2,pass,9
 nvidia_deeprecommender,pass,9
 phlippe_densenet,pass,9
 phlippe_resnet,pass,9
-pyhpc_isoneutral_mixing,pass,0
 pytorch_CycleGAN_and_pix2pix,pass,9
 pytorch_stargan,pass,9
 pytorch_unet,pass,9

benchmarks/dynamo/common.py

@@ -77,7 +77,6 @@ CI_SKIP = collections.defaultdict(list)
 CI_SKIP[CI("eager", training=False)] = [
     # TorchBench
     "DALLE2_pytorch",  # AttributeError: text_encodings
-    "llama",  # does not support complex32
     # TypeError: pad_center() takes 1 positional argument but 2 were given
     "tacotron2",
     # torchrec_dlrm requires gcc-11, https://github.com/pytorch/benchmark/pull/1427
@@ -165,7 +164,6 @@ CI_SKIP[CI("aot_eager", training=True)] = [
 CI_SKIP[CI("inductor", training=False)] = [
     # TorchBench
     "DALLE2_pytorch",  # AttributeError: text_encodings
-    "llama",  # does not support complex32
     # torchrec_dlrm requires gcc-11, https://github.com/pytorch/benchmark/pull/1427
     "torchrec_dlrm",
     "demucs",  # OOM
@@ -217,7 +215,6 @@ CI_SKIP[CI("inductor", training=False, device="cpu")] = [
     "hf_Bert_large",  # OOM
     "hf_GPT2_large",  # Intermittent failure on CI
     "hf_T5_base",  # OOM
-    "llama",  # does not support complex32
     "mobilenet_v2_quantized_qat",
     "pyhpc_turbulent_kinetic_energy",
     "vision_maskrcnn",