[benchmark] Add more timm models (#165381)

Added following models to timm_models

- [convnextv2_nano.fcmae_ft_in22k_in1k](https://huggingface.co/timm/convnextv2_nano.fcmae_ft_in22k_in1k)
- [vit_base_patch14_dinov2.lvd142m](https://huggingface.co/timm/vit_base_patch14_dinov2.lvd142m)
- [ViT-B-16-SigLIP-i18n-256](https://huggingface.co/timm/ViT-B-16-SigLIP-i18n-256)
- [deit_tiny_patch16_224.fb_in1k](https://huggingface.co/timm/deit_tiny_patch16_224.fb_in1k)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165381
Approved by: https://github.com/BoyuanFeng

benchmarks/dynamo/ci_expected_accuracy/aot_eager_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/aot_eager_timm_training.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,7
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,7
+
+
+
 deit_base_distilled_patch16_224,pass,7
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,7
+
+
+
 dm_nfnet_f0,pass,6
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,6
 
 
 visformer_small,pass,7
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,7
+
+
+
+vit_base_patch16_siglip_256,pass,7

benchmarks/dynamo/ci_expected_accuracy/aot_inductor_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/cpu_aot_inductor_freezing_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_amp_freezing_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_freezing_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamic_aot_eager_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamic_aot_eager_timm_training.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,7
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,7
+
+
+
 deit_base_distilled_patch16_224,pass,7
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,7
+
+
+
 dm_nfnet_f0,pass,6
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,6
 
 
 visformer_small,pass,7
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,7
+
+
+
+vit_base_patch16_siglip_256,pass,7

benchmarks/dynamo/ci_expected_accuracy/dynamic_cpu_inductor_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamic_cpu_max_autotune_inductor_amp_freezing_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamic_inductor_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamic_inductor_timm_training.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,7
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,7
+
+
+
 deit_base_distilled_patch16_224,pass,7
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,7
+
+
+
 dm_nfnet_f0,pass,6
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,6
 
 
 visformer_small,pass,7
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,7
+
+
+
+vit_base_patch16_siglip_256,pass,7

benchmarks/dynamo/ci_expected_accuracy/dynamo_eager_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamo_eager_timm_training.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,7
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,7
+
+
+
 deit_base_distilled_patch16_224,pass,7
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,7
+
+
+
 dm_nfnet_f0,pass,6
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,6
 
 
 visformer_small,pass,7
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,7
+
+
+
+vit_base_patch16_siglip_256,pass,7

benchmarks/dynamo/ci_expected_accuracy/inductor_timm_inference.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,0
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,pass,0
+
+
+
 deit_base_distilled_patch16_224,pass,0
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,0
+
+
+
 dm_nfnet_f0,pass,0
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,0
 
 
 visformer_small,pass,0
+
+
+
+vit_base_patch14_dinov2.lvd142m,pass,0
+
+
+
+vit_base_patch16_siglip_256,pass,0

benchmarks/dynamo/ci_expected_accuracy/inductor_timm_training.csv

@@ -10,10 +10,18 @@ beit_base_patch16_224,pass,7
 
 
 
+convnextv2_nano.fcmae_ft_in22k_in1k,fail_accuracy,7
+
+
+
 deit_base_distilled_patch16_224,pass,7
 
 
 
+deit_tiny_patch16_224.fb_in1k,pass,7
+
+
+
 dm_nfnet_f0,pass,6
 
 
@@ -55,3 +63,11 @@ tf_efficientnet_b0,pass,6
 
 
 visformer_small,pass,7
+
+
+
+vit_base_patch14_dinov2.lvd142m,fail_accuracy,7
+
+
+
+vit_base_patch16_siglip_256,pass,7

benchmarks/dynamo/timm_models.py

@@ -271,8 +271,6 @@ class TimmRunner(BenchmarkRunner):
             memory_format=torch.channels_last if channels_last else None,
         )
 
-        self.num_classes = model.num_classes
-
         data_config = resolve_data_config(
             vars(self._args) if timmversion >= "0.8.0" else self._args,
             model=model,
@@ -302,7 +300,6 @@ class TimmRunner(BenchmarkRunner):
         example_inputs = [
             example_inputs,
         ]
-        self.target = self._gen_target(batch_size, device)
 
         self.loss = torch.nn.CrossEntropyLoss().to(device)
 
@@ -370,11 +367,6 @@ class TimmRunner(BenchmarkRunner):
                 tolerance = 1e-2
         return tolerance, cosine
 
-    def _gen_target(self, batch_size, device):
-        return torch.empty((batch_size,) + (), device=device, dtype=torch.long).random_(
-            self.num_classes
-        )
-
     def compute_loss(self, pred):
         # High loss values make gradient checking harder, as small changes in
         # accumulation order upsets accuracy checks.

benchmarks/dynamo/timm_models_list.txt

@@ -1,6 +1,8 @@
 adv_inception_v3 128
 beit_base_patch16_224 128
+convnextv2_nano.fcmae_ft_in22k_in1k 128
 deit_base_distilled_patch16_224 128
+deit_tiny_patch16_224.fb_in1k 128
 dm_nfnet_f0 128
 ghostnet_100 512
 inception_v3 128
@@ -12,3 +14,5 @@ repvgg_a2 128
 swin_base_patch4_window7_224 128
 tf_efficientnet_b0 128
 visformer_small 128
+vit_base_patch14_dinov2.lvd142m 128
+vit_base_patch16_siglip_256 128

benchmarks/dynamo/timm_models_list_cpu.txt

@@ -1,6 +1,8 @@
 adv_inception_v3,128
 beit_base_patch16_224,64
+convnextv2_nano.fcmae_ft_in22k_in1k,128
 deit_base_distilled_patch16_224,64
+deit_tiny_patch16_224.fb_in1k,128
 dm_nfnet_f0,128
 ghostnet_100,128
 inception_v3,128
@@ -12,3 +14,5 @@ repvgg_a2,128
 swin_base_patch4_window7_224,64
 tf_efficientnet_b0,128
 visformer_small,128
+vit_base_patch14_dinov2.lvd142m,128
+ViT-B-16-SigLIP-i18n-256,128