Fix 'seperate' typo in qwen3/glm video-model docstrings

src/transformers/models/glm46v/modeling_glm46v.py

@@ -172,7 +172,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - GLM46V uses timestamps to seperate each video frame, so the video_grid_thw should also be split too.
+        - GLM46V uses timestamps to separate each video frame, so the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -195,7 +195,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/glm4v/modeling_glm4v.py

@@ -1015,7 +1015,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - GLM4V uses timestamps to seperate each video frame, so the video_grid_thw should also be split too.
+        - GLM4V uses timestamps to separate each video frame, so the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -1038,7 +1038,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/glm4v/modular_glm4v.py

@@ -884,7 +884,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - GLM4V uses timestamps to seperate each video frame, so the video_grid_thw should also be split too.
+        - GLM4V uses timestamps to separate each video frame, so the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -907,7 +907,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/glm4v_moe/modeling_glm4v_moe.py

@@ -1184,7 +1184,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - GLM4V_MOE uses timestamps to seperate each video frame, so the video_grid_thw should also be split too.
+        - GLM4V_MOE uses timestamps to separate each video frame, so the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -1207,7 +1207,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/glm_ocr/modeling_glm_ocr.py

@@ -931,7 +931,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - GLM_OCR uses timestamps to seperate each video frame, so the video_grid_thw should also be split too.
+        - GLM_OCR uses timestamps to separate each video frame, so the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -954,7 +954,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/qwen3_5/modeling_qwen3_5.py

@@ -1396,7 +1396,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - Since Qwen3.5 use timestamps to seperate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
+        - Since Qwen3.5 use timestamps to separate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -1419,7 +1419,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py

@@ -1521,7 +1521,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - Since Qwen3.5 use timestamps to seperate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
+        - Since Qwen3.5 use timestamps to separate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -1544,7 +1544,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/qwen3_vl/modeling_qwen3_vl.py

@@ -1041,7 +1041,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - Since Qwen3.5 use timestamps to seperate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
+        - Since Qwen3.5 use timestamps to separate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -1064,7 +1064,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/qwen3_vl/modular_qwen3_vl.py

@@ -741,7 +741,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - Since Qwen3.5 use timestamps to seperate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
+        - Since Qwen3.5 use timestamps to separate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -764,7 +764,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1

src/transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py

@@ -1170,7 +1170,7 @@ def get_rope_index(
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Difference from Qwen2VL/Qwen2.5VL's get_rope_index:
-        - Since Qwen3.5 use timestamps to seperate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
+        - Since Qwen3.5 use timestamps to separate videos, like <t1> <vision_start> <frame1> <vision_end> <t2> <vision_start> <frame2> <vision_end>, the video_grid_thw should also be split too.
 
         Args:
             input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -1193,7 +1193,7 @@ def get_rope_index(
             mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
         """
 
-        # Separate video grid thw into multiple grids because timestamps are used to seperate videos.
+        # Separate video grid thw into multiple grids because timestamps are used to separate videos.
         if video_grid_thw is not None:
             video_grid_thw = torch.repeat_interleave(video_grid_thw, video_grid_thw[:, 0], dim=0)
             video_grid_thw[:, 0] = 1