[V0 Deprecation] Remove unused `context_len` and `seq_len` from M-RoPE (#28395)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/model_executor/models/ernie45_vl.py

@@ -1435,8 +1435,6 @@ class Ernie4_5_VLMoeForConditionalGeneration(
         hf_config: PretrainedConfig,
         image_grid_thw: list[list[int]] | torch.Tensor,
         video_grid_thw: list[list[int]] | torch.Tensor,
-        context_len: int = 0,
-        seq_len: int | None = None,
         second_per_grid_ts: list[float] | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
@@ -1569,7 +1567,6 @@ class Ernie4_5_VLMoeForConditionalGeneration(
             llm_pos_ids_list.append(torch.arange(text_len).view(1, -1).expand(3, -1))
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
-        llm_positions = llm_positions[:, context_len:seq_len]
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
         return llm_positions, mrope_position_delta
 

vllm/model_executor/models/glm4_1v.py

@@ -1622,8 +1622,6 @@ class Glm4vForConditionalGeneration(
         image_grid_thw: list[list[int]] | torch.Tensor | None,
         video_grid_thw: list[list[int]] | torch.Tensor | None,
         second_per_grid_ts: list[float] | None = None,
-        context_len: int = 0,
-        seq_len: int | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
     ) -> tuple[torch.Tensor, int]:
@@ -1754,7 +1752,6 @@ class Glm4vForConditionalGeneration(
             llm_pos_ids_list.append(torch.arange(text_len).view(1, -1).expand(3, -1))
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
-        llm_positions = llm_positions[:, context_len:seq_len]
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
         return llm_positions, mrope_position_delta
 

vllm/model_executor/models/glm4v.py

@@ -625,8 +625,6 @@ class GLM4VForCausalLM(
         hf_config: PretrainedConfig,
         image_grid_thw: list[list[int]] | torch.Tensor,
         video_grid_thw: list[list[int]] | torch.Tensor,
-        context_len: int = 0,
-        seq_len: int | None = None,
         second_per_grid_ts: list[float] | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
@@ -758,7 +756,6 @@ class GLM4VForCausalLM(
             llm_pos_ids_list.append(torch.arange(text_len).view(1, -1).expand(3, -1))
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
-        llm_positions = llm_positions[:, context_len:seq_len]
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
         return llm_positions, mrope_position_delta
 

vllm/model_executor/models/interfaces.py

@@ -995,8 +995,6 @@ class SupportsMRoPE(Protocol):
         image_grid_thw: list[list[int]] | torch.Tensor | None,
         video_grid_thw: list[list[int]] | torch.Tensor | None,
         second_per_grid_ts: list[float] | None = None,
-        context_len: int = 0,
-        seq_len: int | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
     ) -> tuple[torch.Tensor, int]:
@@ -1012,8 +1010,6 @@ class SupportsMRoPE(Protocol):
             image_grid_thw: Image grid dimensions (t, h, w)
             video_grid_thw: Video grid dimensions (t, h, w)
             second_per_grid_ts: Seconds per grid timestep for videos
-            context_len: Context length
-            seq_len: Sequence length
             audio_feature_lengths: Audio feature lengths for multimodal models
             use_audio_in_video: Whether to use audio in video for interleaving
 

vllm/model_executor/models/keye.py

@@ -1630,8 +1630,6 @@ class KeyeForConditionalGeneration(
         hf_config: PretrainedConfig,
         image_grid_thw: list[list[int]] | torch.Tensor,
         video_grid_thw: list[list[int]] | torch.Tensor,
-        context_len: int = 0,
-        seq_len: int | None = None,
         second_per_grid_ts: list[float] | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
@@ -1759,6 +1757,5 @@ class KeyeForConditionalGeneration(
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
-        llm_positions = llm_positions[:, context_len:seq_len]
 
         return llm_positions, mrope_position_delta

vllm/model_executor/models/keye_vl1_5.py

@@ -600,8 +600,6 @@ class KeyeVL1_5ForConditionalGeneration(
         hf_config: PretrainedConfig,
         image_grid_thw: list[list[int]] | torch.Tensor,
         video_grid_thw: list[list[int]] | torch.Tensor,
-        context_len: int = 0,
-        seq_len: int | None = None,
         second_per_grid_ts: list[float] | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
@@ -729,6 +727,5 @@ class KeyeVL1_5ForConditionalGeneration(
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
-        llm_positions = llm_positions[:, context_len:seq_len]
 
         return llm_positions, mrope_position_delta

vllm/model_executor/models/paddleocr_vl.py

@@ -1179,8 +1179,6 @@ class PaddleOCRVLForConditionalGeneration(nn.Module, SupportsMultiModal, Support
         image_grid_thw: list[list[int]] | torch.Tensor,
         video_grid_thw: list[list[int]] | torch.Tensor,
         second_per_grid_ts: list[float],
-        context_len: int = 0,
-        seq_len: int | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
     ) -> tuple[torch.Tensor, int]:
@@ -1293,7 +1291,6 @@ class PaddleOCRVLForConditionalGeneration(nn.Module, SupportsMultiModal, Support
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
-        llm_positions = llm_positions[:, context_len:seq_len]
 
         return llm_positions, mrope_position_delta
 

vllm/model_executor/models/qwen2_5_omni_thinker.py

@@ -927,8 +927,6 @@ class Qwen2_5OmniThinkerForConditionalGeneration(
         image_grid_thw: list[list[int]] | torch.Tensor,
         video_grid_thw: list[list[int]] | torch.Tensor,
         second_per_grid_ts: list[float] | None = None,
-        context_len: int = 0,
-        seq_len: int | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
     ) -> tuple[torch.Tensor, int]:
@@ -1125,7 +1123,6 @@ class Qwen2_5OmniThinkerForConditionalGeneration(
         mrope_position_delta = (
             torch.cat(llm_pos_ids_list, dim=1).max() + 1 - len(src_item)
         )
-        llm_positions = llm_positions[:, context_len:seq_len]
 
         return llm_positions, mrope_position_delta
 

vllm/model_executor/models/qwen2_5_vl.py

@@ -1118,8 +1118,6 @@ class Qwen2_5_VLForConditionalGeneration(
         image_grid_thw: list[list[int]] | torch.Tensor,
         video_grid_thw: list[list[int]] | torch.Tensor,
         second_per_grid_ts: list[float],
-        context_len: int = 0,
-        seq_len: int | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
     ) -> tuple[torch.Tensor, int]:
@@ -1232,7 +1230,6 @@ class Qwen2_5_VLForConditionalGeneration(
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
-        llm_positions = llm_positions[:, context_len:seq_len]
 
         return llm_positions, mrope_position_delta
 

vllm/model_executor/models/qwen2_vl.py

@@ -1240,8 +1240,6 @@ class Qwen2VLForConditionalGeneration(
         image_grid_thw: list[list[int]] | torch.Tensor | None,
         video_grid_thw: list[list[int]] | torch.Tensor | None,
         second_per_grid_ts: list[float] | None = None,
-        context_len: int = 0,
-        seq_len: int | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
     ) -> tuple[torch.Tensor, int]:
@@ -1360,7 +1358,6 @@ class Qwen2VLForConditionalGeneration(
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
-        llm_positions = llm_positions[:, context_len:seq_len]
 
         return llm_positions, mrope_position_delta
 

vllm/model_executor/models/qwen3_omni_moe_thinker.py

@@ -1417,8 +1417,6 @@ class Qwen3OmniMoeThinkerForConditionalGeneration(
         image_grid_thw: list[list[int]] | torch.Tensor | None,
         video_grid_thw: list[list[int]] | torch.Tensor | None,
         second_per_grid_ts: list[float] | None = None,
-        context_len: int = 0,
-        seq_len: int | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
     ) -> tuple[torch.Tensor, int]:

vllm/model_executor/models/qwen3_vl.py

@@ -1419,8 +1419,6 @@ class Qwen3VLForConditionalGeneration(
         hf_config: PretrainedConfig,
         image_grid_thw: list[list[int]] | torch.Tensor,
         video_grid_thw: list[list[int]] | torch.Tensor,
-        context_len: int = 0,
-        seq_len: int | None = None,
         second_per_grid_ts: list[float] | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
@@ -1519,7 +1517,7 @@ class Qwen3VLForConditionalGeneration(
 
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
-        llm_positions = llm_positions[:, context_len:seq_len]
+
         return llm_positions, mrope_position_delta
 
     def get_language_model(self) -> torch.nn.Module:

vllm/model_executor/models/transformers/multimodal.py

@@ -371,8 +371,6 @@ class MultiModalMixin(SupportsMultiModal, SupportsMRoPE):
         image_grid_thw: list[list[int]] | torch.Tensor | None,
         video_grid_thw: list[list[int]] | torch.Tensor | None,
         second_per_grid_ts: list[float] | None = None,
-        context_len: int = 0,
-        seq_len: int | None = None,
         audio_feature_lengths: torch.Tensor | None = None,
         use_audio_in_video: bool = False,
     ) -> tuple[torch.Tensor, int]:
@@ -390,7 +388,7 @@ class MultiModalMixin(SupportsMultiModal, SupportsMRoPE):
             video_grid_thw=video_grid_thw,
         )
 
-        mrope_positions = mrope_positions[:, 0, context_len:seq_len]
+        mrope_positions = mrope_positions[:, 0]
         mrope_position_delta = mrope_position_delta[0].item()
 
         return mrope_positions, mrope_position_delta