A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[VLM][Refactor] Remove useless func `get_input_positions` in `MRotaryEmbedding` (#27088)
Signed-off-by: MengqingCao <cmq0113@163.com>
Mengqing Cao committed
Oct 17, 2025 at 17:00 UTC
e20eba753bbced43837aa92f747e6c50ee36ce09
1 file changed
-34
vllm/model_executor/layers/rotary_embedding/mrope.py
-34
@@ -4,7 +4,6 @@
4
5
import numpy as np
6
import torch
7
-from transformers import PretrainedConfig
7
8
from vllm.triton_utils import tl, triton
9
@@ -376,39 +375,6 @@ class MRotaryEmbedding(RotaryEmbedding):
375
) -> tuple[torch.Tensor, torch.Tensor | None]:
376
return self.forward_native(positions, query, key, offsets)
377
379
- @classmethod
380
- def get_input_positions(
381
- cls,
382
- input_tokens: list[int],
383
- hf_config: PretrainedConfig,
384
- image_grid_thw: list[list[int]] | torch.Tensor | None,
385
- video_grid_thw: list[list[int]] | torch.Tensor | None,
386
- second_per_grid_ts: list[float] | None,
387
- context_len: int = 0,
388
- seq_len: int | None = None,
389
- audio_feature_lengths: torch.Tensor | None = None,
390
- use_audio_in_video: bool = False,
391
- ) -> tuple[list[list[int]], int]:
392
- """Get mrope input positions and delta value."""
393
-
394
- image_grid_thw = [] if image_grid_thw is None else image_grid_thw
395
- video_grid_thw = [] if video_grid_thw is None else video_grid_thw
396
- second_per_grid_ts = [] if second_per_grid_ts is None else second_per_grid_ts
397
-
398
- llm_positions, mrope_position_delta = cls.get_input_positions_tensor(
399
- input_tokens=input_tokens,
400
- hf_config=hf_config,
401
- image_grid_thw=image_grid_thw,
402
- video_grid_thw=video_grid_thw,
403
- second_per_grid_ts=second_per_grid_ts,
404
- context_len=context_len,
405
- seq_len=seq_len,
406
- audio_feature_lengths=audio_feature_lengths,
407
- use_audio_in_video=use_audio_in_video,
408
- )
409
-
410
- return llm_positions.tolist(), mrope_position_delta
411
-
378
@staticmethod
379
def get_next_input_positions(
380
mrope_position_delta: int,