frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Perf] Further optimization for Qwen3-VL fast_pos_embed_interpolate (#25347)

Browse Source 
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
Isotr0py
2025-09-22 04:12:45 +08:00
committed by GitHub
parent 1c3ffdbecc
commit af7dfb0d1a
1 changed files with 30 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
vllm/model_executor/models/qwen3_vl.py
View File

@ -405,25 +405,39 @@ class Qwen3_VisionTransformer(nn.Module):
dh = h_idxs - h_floor
dw = w_idxs - w_floor
w00 = ((1 - dh)[:, None] * (1 - dw)[None, :]).reshape(-1)
w01 = ((1 - dh)[:, None] * dw[None, :]).reshape(-1)
w10 = (dh[:, None] * (1 - dw)[None, :]).reshape(-1)
w11 = (dh[:, None] * dw[None, :]).reshape(-1)
# Create meshgrid view for all h, w vars
dh_grid, dw_grid = torch.meshgrid(dh, dw, indexing='ij')
h_floor_grid, w_floor_grid = torch.meshgrid(h_floor,
w_floor,
indexing='ij')
h_ceil_grid, w_ceil_grid = torch.meshgrid(h_ceil,
w_ceil,
indexing='ij')
h_floor_grid_idx = h_floor_grid * num_grid_per_side
h_ceil_grid_idx = h_ceil_grid * num_grid_per_side
idx00 = (h_floor[:, None] * num_grid_per_side +
w_floor[None, :]).reshape(-1)
idx01 = (h_floor[:, None] * num_grid_per_side +
w_ceil[None, :]).reshape(-1)
idx10 = (h_ceil[:, None] * num_grid_per_side +
w_floor[None, :]).reshape(-1)
idx11 = (h_ceil[:, None] * num_grid_per_side +
w_ceil[None, :]).reshape(-1)
# original computation of weights
# w00 = (1 - dh_grid) * (1 - dw_grid)
# w01 = (1 - dh_grid) * dw_grid
# w10 = dh_grid * (1 - dw_grid)
# w11 = dh_grid * dw_grid
# we reuse w11 here to avoid duplicate
# dh_grid * dw_grid computation
w11 = dh_grid * dw_grid
w10 = dh_grid - w11
w01 = dw_grid - w11
w00 = 1 - dh_grid - dw_grid + w11
indices = torch.stack([idx00, idx01, idx10, idx11], dim=0)
idx00 = h_floor_grid_idx + w_floor_grid
idx01 = h_floor_grid_idx + w_ceil_grid
idx10 = h_ceil_grid_idx + w_floor_grid
idx11 = h_ceil_grid_idx + w_ceil_grid
indices = torch.stack([idx00, idx01, idx10, idx11],
dim=0).reshape(4, -1)
weights = torch.stack([w00, w01, w10, w11],
dim=0).to(dtype=self.dtype,
device=self.device)
weights = weights.unsqueeze(-1)
dim=0).reshape(4, -1, 1)
weights = weights.to(dtype=self.dtype, device=self.device)
embeds = self.pos_embed(indices)
weighted_embeds = embeds * weights