[several models] improve readability (#24585)

* [modeling_clip.py] improve readability

* apply to other models

* fix

src/transformers/models/align/modeling_align.py

@@ -1444,7 +1444,7 @@ class AlignModel(AlignPreTrainedModel):
         self.vision_model = AlignVisionModel(vision_config)
 
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim)
-        self.temperature = nn.Parameter(torch.ones([]) * self.config.temperature_init_value)
+        self.temperature = nn.Parameter(torch.tensor(self.config.temperature_init_value))
 
         # Initialize weights and apply final processing
         self.post_init()

src/transformers/models/altclip/modeling_altclip.py

@@ -1506,7 +1506,7 @@ class AltCLIPModel(AltCLIPPreTrainedModel):
 
         self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
         # Initialize weights and apply final processing
         self.post_init()

src/transformers/models/blip/modeling_blip.py

@@ -743,7 +743,7 @@ class BlipModel(BlipPreTrainedModel):
 
         self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
         # Initialize weights and apply final processing
         self.post_init()

src/transformers/models/bridgetower/modeling_bridgetower.py

@@ -1778,7 +1778,7 @@ class BridgeTowerForContrastiveLearning(BridgeTowerPreTrainedModel):
         self.itc_image_head = BridgeTowerContrastiveHead(config.hidden_size, config.contrastive_hidden_size)
         self.itc_cross_modal_head = BridgeTowerContrastiveHead(config.hidden_size * 2, config.contrastive_hidden_size)
 
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
         # Initialize weights and apply final processing
         self.post_init()
 

src/transformers/models/chinese_clip/modeling_chinese_clip.py

@@ -1376,7 +1376,7 @@ class ChineseCLIPModel(ChineseCLIPPreTrainedModel):
 
         self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
         # Initialize weights and apply final processing
         self.post_init()

src/transformers/models/clap/modeling_clap.py

@@ -1956,8 +1956,8 @@ class ClapModel(ClapPreTrainedModel):
         text_config = config.text_config
         audio_config = config.audio_config
 
-        self.logit_scale_a = nn.Parameter(torch.ones([]) * np.log(config.logit_scale_init_value))
-        self.logit_scale_t = nn.Parameter(torch.ones([]) * np.log(config.logit_scale_init_value))
+        self.logit_scale_a = nn.Parameter(torch.tensor(np.log(config.logit_scale_init_value)))
+        self.logit_scale_t = nn.Parameter(torch.tensor(np.log(config.logit_scale_init_value)))
 
         self.projection_dim = config.projection_dim
 

src/transformers/models/clip/modeling_clip.py

@@ -977,7 +977,7 @@ class CLIPModel(CLIPPreTrainedModel):
 
         self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
         # Initialize weights and apply final processing
         self.post_init()

src/transformers/models/clipseg/modeling_clipseg.py

@@ -979,7 +979,7 @@ class CLIPSegModel(CLIPSegPreTrainedModel):
 
         self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
         # Initialize weights and apply final processing
         self.post_init()

src/transformers/models/flava/modeling_flava.py

@@ -1229,7 +1229,7 @@ class FlavaModel(FlavaPreTrainedModel):
 
         self.image_projection = nn.Linear(self.image_hidden_size, self.projection_dim)
         self.text_projection = nn.Linear(self.text_hidden_size, self.projection_dim)
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
         self.image_to_mm_projection = nn.Linear(self.image_hidden_size, self.mm_hidden_size)
         self.text_to_mm_projection = nn.Linear(self.text_hidden_size, self.mm_hidden_size)

src/transformers/models/groupvit/modeling_groupvit.py

@@ -1368,7 +1368,7 @@ class GroupViTModel(GroupViTPreTrainedModel):
             nn.ReLU(inplace=True),
             nn.Linear(self.projection_intermediate_dim, self.projection_dim, bias=True),
         )
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
         # Initialize weights and apply final processing
         self.post_init()

src/transformers/models/oneformer/modeling_oneformer.py

@@ -399,7 +399,7 @@ class OneFormerLoss(nn.Module):
         self.importance_sample_ratio = importance_sample_ratio
         self.contrastive_temperature = contrastive_temperature
         if self.contrastive_temperature is not None:
-            self.logit_scale = nn.Parameter(torch.ones([]) * np.log(1 / contrastive_temperature))
+            self.logit_scale = nn.Parameter(torch.tensor(np.log(1 / contrastive_temperature)))
 
     def _max_by_axis(self, the_list: List[List[int]]) -> List[int]:
         maxes = the_list[0]

src/transformers/models/owlvit/modeling_owlvit.py

@@ -1065,7 +1065,7 @@ class OwlViTModel(OwlViTPreTrainedModel):
 
         self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
-        self.logit_scale = nn.Parameter(torch.ones([]) * config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(config.logit_scale_init_value))
 
         # Initialize weights and apply final processing
         self.post_init()

src/transformers/models/vision_text_dual_encoder/modeling_vision_text_dual_encoder.py

@@ -204,7 +204,7 @@ class VisionTextDualEncoderModel(PreTrainedModel):
 
         self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
     @add_start_docstrings_to_model_forward(VISION_TEXT_DUAL_ENCODER_TEXT_INPUTS_DOCSTRING)
     def get_text_features(

src/transformers/models/x_clip/modeling_x_clip.py

@@ -1309,7 +1309,7 @@ class XCLIPModel(XCLIPPreTrainedModel):
 
         self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
         self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
-        self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
+        self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
 
         self.prompts_visual_layernorm = nn.LayerNorm(self.vision_embed_dim, eps=config.vision_config.layer_norm_eps)
         self.prompts_visual_projection = nn.Parameter(torch.randn(self.vision_embed_dim, self.projection_dim))